The surface waters of the Northwest Atlantic Ocean are among the fastest warming on Earth. This area is highly-productive biologically, and there are concerns that ecological consequences will follow this rapid-warming. Research on the impacts of this rapid warming has primarily focused on high-profile and/or upper trophic level species. Ecological theory and laboratory studies suggest that elevated temperatures facilitate early maturation and smaller adult body-sizes. However, it is unclear whether that relationship might be mitigated against through adaptive behaviors in an open ocean environment. Here we’ve investigated ecosystem wide impacts on the individual size distribution (ISD) to track changes in community size structure. In cases where community responses are not adequate to counter the impacts of elevated temperatures, we anticipated a steepening of the size spectrum slope (ISD exponent). A steeper relationship relating to a reduction in larger sized individuals and an increased prevalence of smaller sized individuals. Using data from fisheries independent surveys we calculated the community size spectra for four regions along the US NE continental shelf. Correlation/regression analyses were then performed to then assess the degree to which these changes were in alignment to hypothesized bottom-up and top-down disturbances. At the regional scale, we found that community size structure changes (spectra slope) were the largest in the Northern regions, in the Gulf of Maine and Georges Bank. These areas are home to the coldest temperatures and the largest proportions of groundfish species in the community. Spectrum slope declines were most pronounced in the 80’s and 90’s, before the rapid warming of the last decade. The timing of these declines suggest that external factors drove the initial declines of larger-sized individuals within the communities, before elevated temperatures began to influence the ecosystem. Correlation analyses reveal that while fisheries landings are strongly correlated with these declines, bottom-up factors of zooplankton community metrics, Gulf Stream Index, and SST anomalies are also important. While the primary pressure of fisheries exploitation has declined dramatically over time, the recovery of larger-sized individuals has not been seen. That kind of recovery will likely depend on the elevated temperatures seen over the last decade.
Introduction
Understanding the impacts of climate change on local and regional scales is a high priority area of study among the international scientific community. The largest determinant of annual climatic timescale variation is the heat exchange between the atmosphere and the ocean Glantz et al. (1991). Major atmospheric and oceanographic climate modes like the El Nino Southern Oscillation (ENSO), the Gulf Stream Index (GSI) or the North Atlantic Oscillation (NAO) can be linked to localized climate impacts via energy transport between the ocean and atmosphere in a linkage referred to as a “teleconnection” Liu and Alexander (2007). The existence of these pathways demands that scientists understand the inter-connected relationship between local weather and large-scale oceanographic regimes. Scientists following climate change through study of Earth’s energy imbalance have observed that over 90% of that excess heat from climate change is accumulating in the ocean Meyssignac et al. (2019). This fact has re-focused attention on concerns for the future health of marine ecosystems as temperatures, acidification, sea level, and ocean circulation patterns change (“Warming trends increasingly dominate global ocean | nature climate change” (n.d.); Findlay and Turley (2021); University of South Carolina et al. (2021); Frederikse et al. (2020); Neto et al. (2021)). Focusing more narrowly on the impacts of rapid temperature change, scientists have predicted that ecological communities will respond in a number of predictable pathways, each to the limit that their constituent species’ life histories can support a response. These responses include shifts in geographic distribution annd alterations to their growth and maturation (Bergmanns Rule, Blackburn et al. (1999), CITE early maturation when warm, Atkinson (1995) ) . Marine ecosystems are also shaped directly through human interactions like fishing, which has also been shown to impact these single-species characteristics ( CITE: genetic impacts of size-selective gear, allee effects).
Marine ecosystems are complex and are difficult to sample accurately. Much research on the early impacts of climate change on marine systems has focused on individual species or small communities that can be consistently sampled and/or have their interactions explicitly described ( Pinsky et al. (2013), Nye et al. (2009), Pershing et al. (2015) ). An obvious weakness of these approaches is the assumption that all ecological needs provided through inter-species interactions will be equally met under projected climate change conditions. Another lens with which to study the organization of a community is through the study of traits. Size is commonly referenced as the master trait, with well documented relationships with metabolism and bio-energetics Sheldon et al. (1972)Brown et al. (2000)White et al. (2007) . Marine communities are very-often size-structured, with size correlating to trophic position, making them ideal for study through this lens. Size spectra theory is an ecological framework for studying the relationship between the abundance and size of individuals in a community . This relationship relates how efficiently energy flows from the smallest individuals at the lowest trophic levels to larger individuals occupying higher trophic levels Andersen and Beyer (2006)Hillaert et al. (2018) . It is a taxon-agnostic method for detecting changes to ecosystem health and has shown to be sensitive to both large-scale environmental disturbance as well as direct human interaction Shin et al. (2005)Kerr and Dickie (2001) . When organized on a log scale, this relationship between abundance and size reveals a linear decline in abundance with increasing body size White et al. (2008) . Highly productive and diverse marine ecosystems are associated with less steep declines in abundance with size, and support larger numbers of higher trophic level individuals. By studying how this relationship changes through time scientists can observe the energy-transfer efficiency of a community as a signal of ecological health and robustness to external changes in the environment Sprules and Barth (2016) .
Our study is focused on changes in the size distribution finfish community of the Northeast US continental shelf. This is a region that is experiencing rapid increases to temperature tied to changes in the behaviors of the Gulf Stream and Labrador Currents University of South Carolina et al. (2021) . This ecosystem has been sampled extensively as part of national fisheries management efforts and has long-term records on the size distribution of the finish community, providing a long-history with which to study changes in size structure. Based on early observational studies and ecological theory we anticipate that individual level responses to a rapidly warming environment will be detectable in the size-structure of the region’s ecological community, evidence that ecosystems are experiencing change-of-function as a result of rapid environmental change.
Rapid Warming of Northeast Shelf
The Northwest Atlantic is one of the fastest warming locations in the global oceans. Sea surface temperatures in the Gulf of Maine since 1982 have been warming at rates faster than 96% of the world’s oceans, with similar warming rates along the northwest Atlantic shelf (Pershing et al. (2018)). The persistent elevated temperature regime of the area is a result of several forces: a combination of shifting ocean currents and the unique bathymetry of the region. A Northward shift in the Gulf Stream directly increased the regional temperatures through increased transport of warm Gulf Stream water into areas like the Gulf of Maine. The Northward Gulf Stream shift is associated with a higher frequency of warm core rings, and the obstruction of cold-water Scotian Shelf current flow that otherwise counters the influence of the Gulf Stream on the region’s temperatures (Gangopadhyay et al. (2019); Meyer-Gutbrod et al. (2021)). The combination of these oceanographic changes has led to a warmer continental shelf habitat, which has caused alarm among scientists studying the ecology of the region.
Temperature plays a critical role on biological function, affecting many of the chemical reactions that underpin basic physiological function. Reactions impacting critical actions like locomotion and feeding behavior, metabolism and development, and even maturation rates. Because of these relationships, species have evolved thermal preferences around which these functions each operate efficiently. Individuals that are unable to maintain their thermal preferences internally risk metabolic costs unless they can follow their thermal preference in the environment through locomotion or adapt to less-favorable conditions through changes in behavior. In an era of anthropogenic climate change, there is an expectation that many species will be displaced from historic habitats in their efforts to follow their thermal preferences. Recent research in marine environments has shown evidence of this, as species are now shifting to higher latitudes and to deeper depths in the pursuit of more favorable conditions (Nye et al. (2009); Pinsky et al. (2013)). Other work suggests that temperature related impacts may manifest through physiological changes and changes in seasonal phenology, hindering species conservation efforts (Miller et al. (2018); Pershing et al. (2015); Meyer-Gutbrod et al. (2021)). Need to connect temperature to size here: Direct quote from Guiet et al. (2016) , but nails the connection back to temp expectations:
Because it controls chemical reactions, temperature controls metabolic rates which underpin maintenance, growth or reproduction (Clarke and Johnston, 1999; Kooijman, 2010) as well as the functional responses to food density (Rall et al., 2012). Guiet et al. (2016)… In addition to the impact of temperature on communities’ intercepts (heights), the impact of temperature on the speed of the energy flow within communities may affect other properties, such as their resilience to perturbations or the intensity of trophic cascades (Andersen and Pedersen, 2009).
The potential for elevated temperatures to impact the size structure of an ecosystem has implications for the ecosystem resilience in the face of climate change, as well as the blue economies & natural resource systems that rely upon their good health.
Size Spectra Theory
Size is a defining characteristic of species and mediates many ecological interactions and metabolic pathways (Brown et al. 2000) . Size is a major determining factor for the mobility of an organism which directly impacts the ability to evade predation, find foraging success, and efforts to locate and follow seasonal habitats. Size and physical dimensions also impact the metabolic costs associated with each of these behaviors (Hillaert et al. 2018), mediating exchange rates with the immediate environment like heat loss or desiccation in terrestrial species (Gillooly et al. (2001); Heatwole et al. 1969). Body size even informs life history features like life span and the trophic position an individual might occupy through its impact on metabolism and resource use (White et al. 2007). Size structured environments are a fundamental organizational pattern that emerges from these relationships add_citation. Within strongly size-structured ecosystems, growth and maturity changes alter fitness and ultimately determine whether a species is successful in that environment add_citation . Ecological theory is rich with models relating how energy transfers from smaller prey species to larger predatory trophic levels, the allocation of energy for growth, and the trade offs of allocating energy towards those ends (Bertalanffy (1938); Bertalanffy (1957)); add_size_theory_citations). A globally persistent pattern in ecology entangled in those relationships and their critiques is the decline in abundance with increasing body size (Damuth (1981); Currie (1993); Sheldon et al. (1972); Loeuille and Loreau (2006)). The relationship between size and abundance integrates processes operating on the cellular, individual, and community levels simultaneously. By measuring how this relationship changes, scientists can detect whether the fundamental energy transfer pathways from small individuals to larger predators have been impacted or altered, an outward indication of ecosystem-wide changes. The quantities for size and abundance are often the most readily collected data assets of any ecological community. This creates an opportunity to learn much about a system from a relatively low-effort in data collection. For these reasons, size spectrum analyses and individual size distribution (ISD) methods have gained increasing attention as an entry point to assessing ecosystem health and to detect system-wide disturbance (Shin et al. (2005)‘; Pomeranz et al. (2022); White et al. (2007)). Spectra models are also taxon agnostic and avoid the need to explicitly articulate each predator-prey interaction. The “size spectrum” describes the distribution of abundance or biomass as a function of individuals’ mass on a log–log scale (Guiet et al. (2016); Kerr and Dickie (2001)) . Size spectra condense the complexities of predator prey networks and their interactions into taxon-agnostic size-based indices. These indices capture the emergent properties of a system, and have become increasingly used as indicators of ecosystem health. Within the context of fisheries management, changes in spectrum slopes have been associated with fishing exploitation, primarily through the targeted removal of larger individuals (Bianchi et al. (2000); Shin et al. (2005)). Numerical experiments have also linked changes in slope to environmental disturbances (Guiet et al. (2016)). Size spectra have also been shown to express predictable relationships between ecosystems of similar productivity levels as well as from distinct temperature regimes (Guiet et al. (2016)). Use Sprules and Barth Paper to discuss applications Use Pomerantz paper & Edwards to extend into ISD
External Drivers of Size Structure
Our study area’s ecology is one that has historically been studied through the fishing grounds it once supported and their challenges. The focus has traditionally been on the success/failure of single species management actions, however there are growing efforts to better understand the full ecological impact of this endeavor. Fishing practices are often size-selective, with harvesters targeting larger individuals for a higher yield on any time and capital invested. Larger individuals are also commonly the older individuals in a population, having lived longer to achieve those sizes. With the unfortunate side-effect of having an outsized impact on the reproductive potential of the population. Larger individuals have a greater impact on population resilience and recovery, capable of holding more (and often of higher quality) eggs which increase the odds of recruitment success. Heavily fished populations have been shown to exhibit changes to growth rates, suggesting an additional genetic impact from size-selective fishing on the target-species. The direct removal of larger individuals carries a predictable impact on the size distributions of the target species, but may also change that of non-target species as well. Non-target species may be impacted as by-catch from the fishery, or may alternatively find relief from predation if the target species is a direct predator. This can even create circumstances where prey species can crowd out, or outcompete the juveniles of their natural predators, reinforcing the new community. Industrial fishing practices are inherently size-selective, with an outsized footprint relative to small-scale fisheries. Large removals have an immediate and measurable impact on the community size-distribution with potential additional impacts on the future population as well. Size-based harvest in fisheries has been shown to create selective pressures that promote characteristics of early maturation at smaller sizes add_citation. These consequences may independently or in combination create circumstances where a community is unable to return to conditions before these disturbances occurred. reference gb/ss spectra early work (Duplisea & Kerr 1995, or Kerr book)
The ecological community in our study area has with no uncertainty been altered through human behaviors over the last century. Early research estimated that biomass had more than halved in some areas by the 1960’s, pre-dating federal monitoring efforts and inspiring their formation (Fogarty and Murawski (1998)). Key stocks that supported international fishing efforts collapsed, and recovery efforts have in many species fallen short. Work exploring retrospective patterns in stock assessment predictions have highlighted that these shortfalls can be partially attributed to un-accounted for influences of a changing environment. However, other research has noted that the community structure itself may have changed, and with it, how it responds to outside pressure. That same early research noted that species replacement of commercial target species by skate and dogfish was happening as their numbers quickly began to rise. A wealth of anecdotal evidence and records from commercial landings support the conclusion that the region does not support either the fisheries or the fishing communities in the same way today. The Lobster industry remains King in the Gulf of Maine, another side-effect of the decline in large predators in the area. With other similar declines in large-fish populations up and down the coast, scientists have recently brought the focus of their attention away from single-species management to more holistic and ecosystem-wide approaches.
Regime Shifts in Marine Systems
Goals:
Want to hit: dynamic tipping points, hysteresis, and non-stationary functional relationships. Show examples of the idea on single species, suggest that we size spectra can be used to check the whole community for tipping points
When an ecosystem experiences an abrupt change in its internal dynamics in response to internal or external pressures, it is said to have crossed a tipping point. When these tipping points are crossed, it often becomes difficult or impossible to return to the previous state. This is true even when the pressures that caused the shift have ceded and favorable conditions have returned, a phenomenon known as hysteresis. This phenomenon has been observed in individual population trends, and is a major factor undermining sustainable management of fisheries (Blocker 2023; Sguottti 2019) . Another feature that may accompany a regime shift is the altering of functional relationships with other biotic or abiotic features of the environment. When this occurs, the productivity of a species and/or its resilience to stress may be altered, and assumptions underpinning sustainable harvest thresholds should be updated (Blocker 2023). At the community level the alternative community states may support a different composition of species. The community itself may respond differently to the same stressors/stimulus. In the Northeast Shelf Region there is evidence to suggest that through a combination of human influences and environmental changes that the ecology of some areas may have crossed ecological tipping points. This study explores whether evidence for an ecological tipping point can be seen through the perspective of the size spectrum, and whether there is evidence for hysteresis, or non-stationarity in the community’s relationships to external forces.
Purpose
Understanding that human populations depend on the health of their ecosystems, there is a need to better understand when we’ve crossed ecological tipping points. One way to capture these changes is through the use of community-wide metrics that capture the inner-efficiencies of many trophic interactions. Size based indices are metrics that can be estimated from the information that has historically been available from long-term survey efforts. These indices have been shown to be sensitive to the impacts of fishing, but also capture environmental dynamics as well. We estimated size spectrum relationships to use as size based indicators of the ecological communities within each sub-region of the Northeast US continental shelf. Based on ecological theory we believe that the sustained increases in temperature in the NW Atlantics should have a physiological impact on the community size structure. We hypothesize that rapid warming alters the community through the direct influence of temperature on metabolism, growth, and population productivity. We also hypothesize that size spectra will uniquely reflect the physical environment of the regions, with less steep size spectra in regions with colder temperatures and of higher primary productivity. Understanding that temperature is not the sole mechanism for altering community size structure, we will explore how the temporal structures in size spectra have changed in relation to major external drivers. We posit that distinct patterns of deviation from that signature may reflect a fundamental change in the ecology of that region, an indication of a possible ecological regime change, and an altered relationship to human and environmental disturbances.
Methods
Fish Data Source and Processing
Data on the biomass, abundance, and size of fish on the Northeast U.S. Shelf were collected as part of the Northeast Fisheries Science Center’s bottom trawl survey (Grosslein 1969, Azarovits 1981, Politis 2014). This survey is conducted from Cape Hatteras, North Carolina to the Gulf of Maine each year in the spring and in the fall. The survey follows a stratified random sampling design, with strata defined based on depth, bottom habitat, and latitude. Trawls are performed for a fixed duration at each station, reporting aggregate abundance and biomass for all species caught, and measuring individual lengths and weights for the catch of each species or a sub-sample if that catch is large. Correction factors were applied to aggregate species abundance and biomass to account for changes in vessels, gear, and doors used in the survey over time (Sissenwine and Bowman 1978, Byrne and Forrester 1991, Miller et al. 2010). However, abundance and biomass at length needed to be estimated after these aggregate corrections. As such, abundance at length for each species was adjusted to match the corrected aggregate species abundance at each station, such that for each species, the sum of the resulting estimated abundance numbers across each length is equal to the corrected aggregate abundance.
Community Composition & Functional Groups
Analyses were performed using 68 species. These species were selected based on the availability of published weight-at-length relationships (Wigley et al. 2003) and represented 98.98% of the total biomass caught in the survey. Each species was assigned to a functional group based on life history and geography using the definitions of (Hare et al. (2016)). Functional groups included coastal fishes, diadromous fishes, elasmobranchs, groundfish, pelagic fishes, and reef fish species (Table 1.). Six species with available length-weight details did not have a functional group designation, these species were designated as reef species. Exploratory analyses showed that the pelagic species biomass was low in all regions, and is unlikely to be representative of true biomass trends due to gear selectivity.
Published length-weight relationships ( Wigley et al. 2003 ) were used to convert from length data, available for all individuals, into their corresponding biomass-at-length. To account for differences in sampling effort among survey strata, all corrected abundance-at-length data were area-stratified. Area-stratified biomass-at-length values were then computed as the product of area-stratified abundance-at-length and estimated weight-at-length. All analyses were performed using area-stratified abundances and their associated area-stratified biomass estimates, hereby referred to as simply abundance-at-length & biomass-at-length
Community Metrics
Our analyses used all data collected during the spring and fall surveys from 1970-2019. Data were grouped using survey-design strata into four sub-regions: Gulf of Maine, Georges Bank, Southern New England, Mid-Atlantic Bight (Figure 1.). These sub-regions have been widely used in regional ecological studies (e.g., ). For each region, we developed the following time series of ecological indicators:
Annual mean abundance and biomass by functional group
Annual mean length and weight of the aggregate community and for each functional group
Annual estimates of the community size spectrum slope
Quantifying Body Size Changes
Changes in the size structure of the community was measured using the average length and weight across all species, each weighted by their species and size-specific catch rates. The average body length (cm) and body weight (kg) was calculated for each region and within each functional group.
\[X_j = \frac{ \sum_{}^{} n_iX_{ij} }{ \sum~w_i }\] Where \(X\) is the body-size metric, \(j\) is the year, & \(n\) is the abundance for each station \(i\).
Data for body size trends were not truncated using any minimum or maximum size and reflect all available catch data for the 68 species in this study for which biomass-at-length could be estimated.
Estimating Size Spectra
Abundance size spectra were calculated using the area-weighted abundance at length information from the catch data. Lengths of individuals in the catch data are measured to the nearest cm, with smaller specimens measured to the nearest millimeter. Because individual biomass is estimated from those length measurements, there is a range of possible body mass values between the increments used (cm and mm). The relationship between length and mass in fishes is exponential and taxon specific. Using only the lower or upper end of those ranges introduces biases which are different for each species and increase with larger body-sizes. To account for these biases, we used the extended likelihood method (MLEbin) of Edwards et al. (2020). This method estimates the exponent of size spectra (b) for a bounded power law relationship between abundance and length-estimated biomass. The exponent of this relationship is analogous to slope estimates from linear regression on logarithmic axes, with a steeper slope (i.e. a more negative b) indicating fewer large-bodied individuals and/or more small-bodied fishes (Carvalho 2021) . This method has been shown to be the most accurate for estimating the exponent of size spectra when tested against alternative methods that require arbitrary decisions around binning the data and make estimates difficult to compare directly across studies (White 2008; Sprules & Barth 2016; Edwards 2017).
Using this method, the size spectra exponent (b) was estimated for each region from 1970 to 2019. A minimum biomass of 1g was used for the lower bound and a maximum biomass of 10kg was used as an upper bound for the ISD’s bounded power law probability density function Equation 1, where X is body mass & Λ is the scaling exponent of the abundance density function. This body-size range constitutes 97.83% of all the biomass for the 68 species included in this study. This body-size range truncation was done to account for poor gear selectivity at the smallest and largest size ranges. This measure also imposes shared bounds to the size range covered by our size spectra, reflecting only the ranges we’d expect to consistently sample across these different areas. Exponents of size spectra (b) were calculated using code modified from the sizeSpectra package (Edwards et al. (2017); Edwards et al. (2020)) and implemented using the R statistical programming language.
Drivers of Size Spectra Changes
The impact of external factors on the changes in size spectra were explored using multiple regression analyses. Annual variation in size spectrum slopes was modeled using several hypothesized drivers including the environmental state and measures of anthropogenic disturbances. The large-scale environmental drivers included were regional SST anomalies & the larger-scale impact of the Gulf Stream Index (GSI). Fishing pressure represents the primary top-down anthropogenic driver in the region, and was investigated using the aggregate regional commercial landings data. Bottom-up ecological interactions were explored using zooplankton indices from the EcoMon plankton survey. Each of these drivers (climate, productivity, fishing) have independently been shown in other works to have measurable impacts on size spectra. Our exploratory investigations here are an effort to evaluate which best explain the trends within our study region, and whether those relationships have changed over the study period.
Gulf Stream Index
Data for the Gulf stream index (GSI) was obtained from the ecodata package in R (Bastille & Hardison 2018) . This package supplies GSI data at monthly intervals following the methodology of Pérez-Hernández and Joyce (2014) and Joyce et al. (2019) , using as sea level height anomaly data from the Copernicus Marine Environment Monitoring Service.
Sea Surface Temperature Data
Global sea surface temperature (SST) data were obtained via NOAA’s Optimum Interpolation SST analysis (OISSTv2), which provides daily SST values at a 0.25° latitude x 0.25° longitude resolution (Reynolds et al. 2007). A daily climatology for every 0.25° pixel in the global data set was created using average daily temperatures spanning the period of 1982-2011. Daily anomalies were then computed as the difference between observed temperatures and the daily climatological average. OISSTv2 data used in these analyses were obtained from the NOAA Physical Sciences Laboratory, Boulder, Colorado, USA from their website at https://psl.noaa.gov/data/gridded/data.noaa.oisst.v2.html.
SST data were regionally averaged to match the four survey sub-regions described above (Figure 1a), producing daily time series for each area. These were then averaged into annual time series of surface temperatures and anomalies. All averaging between satellite grid units was done with area-weighting of the grid cells to account for differences in cell size with latitude/longitude in the OISSTv2 data.
Commercial Fishing
Fishing pressure in the region was indexed using state and federal commercial fishing landings. These data were obtained from the Greater Atlantic Regional Fisheries Office (GARFO) for statistical areas that are routinely used for fisheries reporting and management (Figure 1., right). Individual statistical areas were aggregated into regions that closely align with the survey areas we defined for the size spectra analyses (Figure 1 b.). Each region’s commercial landings timeline was scaled using z-score normalization
Zooplankton Indices
Zooplankton data was from the National Oceanographic and Atmospheric Administration Marine Resources Monitoring, Assessment and Prediction (MARMAP) program and Ecosystem Monitoring (EcoMon) cruises detailed extensively in Kane (2007), Kane (2011), and Morse et al. (2017). Abundance anomalies were computed from the expected abundance on the day of sample collection. The small copepod index is computed by averaging the individual abundance anomalies of Pseudocalanus spp., Centropages hamatus, Centropages typicus, and Temora longicornis. The “large-copepod” anomaly values are the abundance anomaly of Calanus finmarchicus, the largest copepod in the Northeast U.S. region. Anomalies for both large and small zooplankton groups were averaged within ecological production units by the data provider, producing annual timeseries for the Gulf of Maine, Georges Bank, and Mid-Atlantic Bight. The timeseries for the Mid-Atlantic Bight EPU was used as a predictor for both the Mid-Atlantic Bight and Southern New England regions in our analyses. This decision was made to address how the spatial coverage of this EPU spans the joint area of our two southernmost regions.
Exploring Influence of Drivers on Size Spectra
The hypothesized drivers for size spectrum changes were evaluated using multiple regression analysis. Candidate models for each region were informed by the any observed temporal structures in the size spectra changes through time, and with the inclusion of important lag-effects of predictors.These modeling decisions were informed by the following steps performed prior to multiple regression analysis with the above drivers.
Temporal Lags in Hypothesized Drivers
In addition to exploring their contemporaneous impacts, the importance of any time-lags on predictors was also evaluated. Important lagged relationships were identified through the use of cross correlation function (CCF) estimates. These estimates looked at correlations at one-year lag intervals to investigate potential time-delayed effects between the hypothesized drivers and the community’s size spectra. CCF were performed between the dependent variables of the size spectra slope with the independent variables of the annual gulf stream index, the corresponding regional commercial landings and sea surface temperature anomalies, and both small and large zooplankton indices. This was performed for annual lags up to 10 years. Significant lags identified in the CCF analysis were then included as additional candidate predictors in the multiple regression analysis of spectrum slope changes.
Non-Stationary Behavior Across Response Regimes
To highlight potential signatures of any non-stationary or non-linear discontinuous dynamics, a hallmark of regime-shift dynamics (Blocker 2023), we explored discontinuous changes in the size spectra using breakpoint analyses. Breakpoint analysis was performed on the annual time series of spectrum slopes to explore their temporal structures for evidence of any potential regime shifts. Breakpoint analyses were performed using the envcpt package (Killick et al. 2021). This package applies an automatic model selection process testing support for candidate model structures including constant/piecewise changes to the mean, variance, trends, & autocorrelations and any identified changepoints located using the pruned exact linear time algorithm (Rikardsen 2004). Best supported model structures from this procedure were then used to inform whether (if any) regional spectra expressed distinct periods of behavior. In the events where changepoints were identified, this was used as evidence that there may have been a community regime change. Following this thinking it is plausible for there to be inconsistencies or non-stationarity in the predictors’ impact on the community spectra, as different community regimes may be differently vulnerable to top-down pressures like commercial fishing. To allow for changes in these relationships across different regimes, an additional interaction term was added to the regression model candidates for that region.
Regime Specific Correlations
Based on any breakpoints detected, single driver correlations were prepared to visually explore how each driver was correlated within the different regimes. Additionally, spectra slope changes were also correlated with the body-size changes of each functional group as a window into what body-size changes might be informing the change in the size structure.
Multiple Regression Modeling of Hypothesized Drivers
Following these explorations into temporal structures, multiple regression models were developed and evaluated independently for each region. Potential drivers included in the models were the annual Gulf Stream Index, the annual regionally averaged SST anomaly value, regionally averaged small and large copepod indices, and the scaled annual commercial landings from that region. Gulf stream index, zooplankton indices, and SST anomalies were not standardized prior to use in model fitting and evaluation as the two former are already an index value and the latter is scaled to a thirty-year climatological reference period. In the event that CCF analysis identified an important time lag for a region’s independent variables, these specific time-lagged predictors were included as predictors in the candidate models, otherwise time-lags on predictors were not evaluated. For any region(s) that showed support for changepoints in their temporal structure, the periods separated by those breaks were included as an interaction term on all other predictors. In the absence of any identified changepoints, no multi-year periods were included as potential regime impacts. All candidate models for a region were evaluated using AIC and AIC weights to rank the most parsimonious models for each region’s dynamics.
Results
Abundance Changes
Stratified abundance estimates have been gradually increasing on the Northeast Shelf since 1970. Abundance growth accelerated after 2007 to peak levels in 2014. Abundance then began declining, which continued through 2019 (Figure 2).
In the Gulf of Maine, fish abundance remained relatively-low until the 1990’s, at which point it began to steadily rise–closely aligning with the pattern for the Northeast Shelf. This increase in abundance reversed briefly during 2002-2006, but continued to rise after 2006 before hitting a regional peak in 2016. Populations then declined through 2019 similar to the shelf-wide trend. Georges Bank abundances were consistently low from 1970 around 2010. By 2014 abundance had roughly quadrupled, propelled by strong recruitment classes of haddock. After 2014, abundance quickly fell to numbers more similar to the 1970-2010 levels by the end of the decade. Community abundance in Southern New England displayed higher inter-annual changes across all years compared to both Gulf of Maine and Georges Bank. Abundance in this area showed a less dramatic rise and fall than the Northern regions. Abundances began increasing rapidly here in 2007, before falling back to earlier levels by the end of the 2010’s decade. The Mid Atlantic Bight displayed the most inter-annual variability and had no major trends in fish abundance.
Groundfish species were the dominant functional group driving the abundance and biomass trends in the Gulf of Maine and Georges Bank, with the two southern regions showing a more balanced abundance distribution among the five functional groups (Figure 2).
Total Abundance by Individual Body-Size
Biomass Changes
Similar to abundance, the overall biomass was highest in the two northern regions, the Gulf of Maine and Georges Bank. Roughly half of the biomass sampled in these regions can be attributed to groundfish species, with the second largest contributions coming from elasmobranchs. Within the groundfish biomass, larger individuals >2kg in particular, declined during the 70’s and 80’s in these regions, never truly recovering. Beginning in the 2000’s there were signs that groundfish abundances were increasing as evidenced by increasing numbers of smaller individuals, however in both regions this trend appears to have reversed by the mid 2010’s. Elasmobranch biomass increased steadily throughout the survey time period across all regions, with the exception of southern New England. This area showed large 5-10 year swings in biomass, but no clear long-term trend. Larger elasmobranch were rare in all regions except for a period spanning the late 70’s through the early 90’s isolated to Georges Bank. Demersal species biomass was highest in the Gulf of Maine, dwarfing their contributions in other regions. Their biomass declined in the 70’s, was flat until the late 90’s, remaining relatively high until declining in the late 2010’s. Pelagic species biomass was low in all regions, and is unlikely to be representative of true biomass trends due to gear selectivity.
Regional Biomass
There was a distinct difference between Northern and Southern regions in the way biomass was distributed among the different functional groups. The largest contributors to biomass in the southern regions (southern New England & mid-Atlantic bight) was the elasmobranch community. While the northern regions (Gulf of Maine & Georges Bank) each had similar quantities of elasmobranch biomasses, there was also a comparable contribution of groundfish and in the Gulf of Maine there was a major component of demersal species as well.
The average fish size in the Gulf of Maine (length and weight) declined the greatest of all regions over our study period. The average individual length was greatest in the 1970’s in the 35-40cm range, falling to 28-33cm over the last decade. Body-weight fell dramatically in the 1980’s, from around .75kg in the 1970’s to .25-.30kg, roughly a third of what it had been. Georges Bank body sizes also declined during the study period, but less dramatically. Both of these Northern regions had brief period in the late 2000’s where average length and weight rose, before falling again in the 2010’s. The MAB region was the only region to see a long-term increase in both length and weight during the study period. SNE saw no long-term change in length, and a minor decline in average body-weight.
At the start of our time series, back in the 1970’s, there was a clear difference in the relative positions of spectra parameters among the different regions. Gulf of Maine and Georges Bank showed the least steep spectra slopes in the earlier time periods with slopes around -1 & -1.1 respectively. The relatively flat slopes in these regions both steepened over time, settling near -1.3 (GoM) and -1.5 (GB). Gulf of Maine experienced much of its decline during the 1980’s and 1990’s. There was a brief reversal in this trend during the 2000’s, but slopes continued to steepen by 2010 and remained steep through 2019. Georges Bank did not experience as rapid of a decline, but experienced a similar long-term steepening. In contrast to the northern regions, SNE and MAB had steeper slopes in the -1.2 to -1.5 territory. The long term pattern for SNE was one of increasing volatility, but not so much a decline. The spectra slope for the MAB was less volatile, but similarly maintained a relatively stable wander around -1.4. By the end of the study period all regions had slopes that were at or near a similar level.
Exploratory analysis on potentially important predictor time lags through the use of cross correlation functions raised 6 potential lagged-driver candidates for the driver regression analyses. CCF identified potential time lags in Georges Bank predictors, and in the Mid-Atlantic Bight. Lagged predictors added to Georges Bank regression model selection process included: A 2-year lag on the small zooplankton index, a 1-year lag on the large zooplankton index, a 1-year lag of SST, 4-year lag & 1-year lags of the GSI, and 4 & 5 year lags of commercial landings. Lagged predictors added to Mid-Atlantic Bight regression model selection process included: 5 & 7 year lags on the small zooplankton index and a 2-year lag on SST.
Spectra Slope Changepoints
Breakpoint analyses showed support for ecological regime change in one region, the Gulf of Maine. Evidence of breakpoints were found in 1999 & 2007 – when slope values increased briefly, before reversing course and falling further. The best-supported breakpoint model structure included breaks in the linear trend at 1999 & 2007, and a 2-year autocorrelation term. Suggesting that the community spectra behaved differently over 3 regimes with strong autocorrelation to its state two years prior. For Georges Bank spectra slopes showed a single trend structure, with slopes steepening over time. This was also the case for the Mid-Atlantic Bight, but with slopes becoming more shallow with time. In Southern New England a single mean (intercept) model best reflected the lack of change. With no strong support for either a long-term trend or break points. Based on these exploratory results, three regime periods were added to the Gulf of Maine’s dataset for multiple regression analysis. The three regimes were 1982-1998, 1999-2006, 2007-2019. These regimes were included as an interaction term with each of the original hypothesized drivers, permitting potentially different influences of the drivers within each regime. The two-year lag on the spectrum slope was also included based on its support in the most-parsimonious changepoint analysis model. No breakpoints or autocorrelation terms were added to models for the other regions.
Region
Most Supported Model
gom
trendar2cpt
gb
trend
sne
meancpt
mab
trend
Change point analyses suggest there were as many as 3 distinct regimes in the Gulf of Maine slope changes. For all other regions there was not strong support for change points.
Spectra Slope Regression Analysis
WORKING HERE - Effect Reporting
fix the coefficients so the interactions now reflect the net result
Model rankings using AICc & delta-AICc (Figure 5.) for the Gulf of Maine’s size spectrum slope models best support a regression model containing two predictors with a stationary effect across all years, and two predictors whose effects across were allowed to vary across the regime periods identified by the changepoint analysis (non-stationary effects). The two stationary drivers impacts were the negative impact of the small zooplankton index (p = 0.011) and a negative impact from the two-year autocorrelation term highlighted by the exploratory changepoint analysis (p = 0.033). The inclusion of an interaction term between the three regimes (1982-1998, 1999-2006, & 2007-2019) and both the commercial landings index and the large zooplankton index allowed these drivers to differently impact size spectrum slope within the three regimes. Over those three periods the impact of landings during the first regime was positive (p = 0.009), the 1999-2006 regime was negative (p = 0.032), and during the third regime there was no relationship (p = 0.088). For the large zooplankton index during the first two regimes there was no significant impact (p = 0.2, p = >0.9), however during the third regime large zooplankton had a positive effect on the spectrum slope (p = <0.001). None of the terms for the regime periods themselves were significant in the final model. Models containing only stationary effects on the predictors were not retained by model selection. This agreed with visual assessments of their predictive performance and residual trends.
The best supported model for Georges Bank retained two predictors: sea surface temperature anomalies (p = 0.028) and the small zooplankton index at a 2-year lag (p = 0.006). With both the SST anomalies and the lagged small zooplankton index having negative effects on size spectrum slope. Five candidate models were within a delta-AIC range of <2. These top-models retained only bottom-up drivers as the best predictors, usually as some pair containing the small zooplankton at 0-2 year lags and either SST anomalies or the Gulf Stream Index. Suggesting that this region’s size spectrum changes are most highly correlated with environmental forces and not commercial landings.
Exploratory analysis for Southern New-England was suggestive of a lack of trend in size spectrum slopes. This was further confirmed by the model selection process. The best supported model of Southern New England retained only commercial landings, however this relationship was not significant (p = 0.13) and had very low performance (r-squared = 0.06).
Three top models were selected for the Mid-atlantic Bight region, all with non-stationary predictors. The best supported model showed that increases in the small zooplankton index had a negative impact on spectrum slope (p = 0.020). This result was present in the other top candidates, which each also included a negative correlation to large zooplankton or with commercial landings. Model performance was low among the top models (r-squared 0.14-0.16).
The following panels show the historical changes in each of the drivers. Landings have been scaled by average total landings within each region across all years. SST and GSI have not been scaled. This is different from how they are implemented in the regression analyses, when the landings were scaled over the 1982-2019 period.
Common and scientific names for the species that constitute each
functional group used in our analyses. X markers are used to indicate
which regions each species has been caught in the data.
Functional Group Assignments and Regional Presence/Absence
Common Name
Scientific Name
Georges Bank
Gulf of Maine
Mid-Atlantic Bight
Southern New England
Coastal - (18)
Atlantic Croaker
micropogonias undulatus
X
X
X
Atlantic Spadefish
chaetodipterus faber
X
Atlantic Thread Herring
opisthonema oglinum
X
X
Black Sea Bass
centropristis striata
X
X
X
X
Blackbelly Rosefish
helicolenus dactylopterus
X
X
X
X
Blueback Herring
alosa aestivalis
X
X
X
X
Bluefish
pomatomus saltatrix
X
X
X
X
Butterfish
peprilus triacanthus
X
X
X
X
Cunner
tautogolabrus adspersus
X
X
X
X
Greater Amberjack
seriola dumerili
X
X
Northern Kingfish
menticirrhus saxatilis
X
X
X
Scup
stenotomus caprinus
X
X
X
X
Southern Kingfish
menticirrhus americanus
X
Spanish Mackerel
scomberomorus maculatus
X
Spanish Sardine
sardinella aurita
X
Spot
leiostomus xanthurus
X
X
Striped Bass
morone saxatilis
X
X
X
X
Weakfish
cynoscion regalis
X
X
X
Diadromous - (2)
American Shad
alosa sapidissima
X
X
X
X
Atlantic Sturgeon
acipenser oxyrhynchus
X
Elasmobranch - (19)
Atlantic Angel Shark
squatina dumeril
X
Atlantic Sharpnose Shark
rhizoprionodon terraenovae
X
Barndoor Skate
dipturus laevis
X
X
X
X
Bullnose Ray
myliobatis freminvillei
X
X
Chain Dogfish
scyliorhinus retifer
X
X
X
Clearnose Skate
raja eglanteria
X
X
Cownose Ray
rhinoptera bonasus
X
Little Skate
leucoraja erinacea
X
X
X
X
Rosette Skate
leucoraja garmani
X
X
X
X
Roughtail Stingray
dasyatis centroura
X
Sand Tiger
carcharias taurus
X
Sandbar Shark
carcharhinus plumbeus
X
X
Smooth Butterfly Ray
gymnura micrura
X
Smooth Dogfish
mustelus canis
X
X
X
X
Smooth Skate
malacoraja senta
X
X
X
X
Spiny Butterfly Ray
gymnura altavela
X
Spiny Dogfish
squalus acanthias
X
X
X
X
Thorny Skate
amblyraja radiata
X
X
X
X
Winter Skate
leucoraja ocellata
X
X
X
X
Groundfish - (25)
Acadian Redfish
sebastes fasciatus
X
X
X
X
American Plaice
hippoglossoides platessoides
X
X
X
X
Atlantic Cod
gadus morhua
X
X
X
X
Atlantic Halibut
hippoglossus hippoglossus
X
X
X
Atlantic Wolffish
anarhichas lupus
X
X
X
Cusk
brosme brosme
X
X
X
X
Fawn Cusk-Eel
lepophidium profundorum
X
X
X
X
Fourspot Flounder
paralichthys oblongus
X
X
X
X
Goosefish
lophius americanus
X
X
X
X
Haddock
melanogrammus aeglefinus
X
X
X
X
Longhorn Sculpin
myoxocephalus octodecemspinosus
X
X
X
X
Northern Searobin
prionotus carolinus
X
X
X
X
Ocean Pout
macrozoarces americanus
X
X
X
X
Offshore Hake
merluccius albidus
X
X
X
X
Pollock
pollachius virens
X
X
X
X
Red Hake
urophycis chuss
X
X
X
X
Sea Raven
hemitripterus americanus
X
X
X
X
Silver Hake
merluccius bilinearis
X
X
X
X
Spotted Hake
urophycis regia
X
X
X
X
Summer Flounder
paralichthys dentatus
X
X
X
X
White Hake
urophycis tenuis
X
X
X
X
Windowpane Flounder
scophthalmus aquosus
X
X
X
X
Winter Flounder
pseudopleuronectes americanus
X
X
X
X
Witch Flounder
glyptocephalus cynoglossus
X
X
X
X
Yellowtail Flounder
limanda ferruginea
X
X
X
X
Pelagic - (4)
Atlantic Herring
clupea harengus
X
X
X
X
Atlantic Mackerel
scomber scombrus
X
X
X
X
Buckler Dory
zenopsis conchifera
X
X
X
X
Round Herring
etrumeus teres
X
X
X
X
Functional group assignments adapted from Hare et al. 2010
Top Commercial Fisheries Landings of Northeastern US (by weight)
Avg. Annual Landings (lb.)
Total Landings (lb.)
Total Value ($)
Gulf of Maine - 1960
Hake, Silver
16.58M
281.87M
8.71M
Herring, Atlantic
11.57M
138.83M
2.50M
Redfish, Acadian
2.12M
88.97M
3.41M
Gulf of Maine - 1970
Herring, Atlantic
22.78M
501.08M
19.70M
Menhaden, Atlantic
17.78M
373.48M
7.87M
Redfish, Acadian
3.14M
219.85M
23.87M
Gulf of Maine - 1980
Herring, Atlantic
21.78M
653.26M
34.52M
Menhaden, Atlantic
21.24M
509.75M
12.77M
Pollock
3.33M
229.57M
62.00M
Gulf of Maine - 1990
Herring, Atlantic
25.21M
958.12M
54.12M
Cod, Atlantic
2.35M
138.76M
131.76M
Shark, Dogfish, Spiny
3.34M
120.17M
15.95M
Gulf of Maine - 2000
Herring, Atlantic
2.99M
47.77M
4.31M
Monkfish/Angler/Goosefish
716.21K
31.51M
51.13M
Cod, Atlantic
692.95K
30.49M
42.30M
Gulf of Maine - 2010
Tuna, Bluefin
209.06K
3.76M
33.30M
Shark, Dogfish, Spiny
479.11K
2.87M
590.62K
Pollock
188.20K
1.69M
2.08M
Georges Bank - 1960
Haddock
15.00M
270.06M
34.41M
Hake, Silver
6.83M
95.57M
3.19M
Cod, Atlantic
4.88M
87.89M
8.12M
Georges Bank - 1970
Cod, Atlantic
7.78M
233.48M
59.16M
Flounder, Yellowtail
4.62M
138.52M
43.16M
Redfish, Acadian
2.63M
76.37M
9.09M
Georges Bank - 1980
Cod, Atlantic
10.11M
404.40M
211.60M
Flounder, Winter
2.50M
100.11M
89.84M
Haddock
2.36M
94.27M
66.68M
Georges Bank - 1990
Cod, Atlantic
4.27M
192.29M
190.26M
Hake, Silver
1.79M
76.82M
20.49M
Flounder, Winter
1.23M
56.43M
75.59M
Georges Bank - 2000
Cod, Atlantic
2.17M
62.91M
75.20M
Herring, Atlantic
3.49M
48.92M
3.73M
Haddock
1.54M
43.01M
55.37M
Georges Bank - 2010
Hake, Silver
155.88K
779.40K
499.90K
Haddock
39.65K
118.95K
143.04K
Flounder, Winter
40.40K
80.80K
216.28K
Southern New England - 1960
Other Fish, Bony
14.84M
400.77M
3.73M
Flounder, Yellowtail
6.56M
196.83M
19.12M
Flounder, Winter
2.52M
70.58M
7.01M
Southern New England - 1970
Menhaden, Atlantic
9.99M
239.84M
5.12M
Other Fish, Bony
4.05M
206.59M
2.49M
Flounder, Yellowtail
2.07M
153.55M
36.47M
Southern New England - 1980
Menhaden, Atlantic
6.60M
217.68M
10.21M
Hake, Silver
2.56M
205.02M
46.11M
Flounder, Yellowtail
1.66M
132.92M
83.38M
Southern New England - 1990
Hake, Silver
2.52M
196.81M
78.54M
Herring, Atlantic
2.12M
129.02M
7.19M
Menhaden, Atlantic
3.71M
125.98M
8.69M
Southern New England - 2000
Mackerel, Atlantic
2.55M
135.06M
15.60M
Hake, Silver
1.00M
55.25M
26.89M
Skate, Nk
950.56K
49.43M
6.53M
Southern New England - 2010
Scup
161.10K
6.44M
4.29M
Hake, Silver
145.07K
4.21M
3.12M
Flounder, Summer
80.43K
3.86M
11.54M
Mid-Atlantic Bight - 1960
Flounder, Summer
2.03K
4.05K
720.00
Flounder, Yellowtail
2.33K
2.33K
214.00
Flounder, Witch
395.00
395.00
36.00
Mid-Atlantic Bight - 1970
Menhaden, Atlantic
10.20M
50.98M
1.59M
Weakfish/Sea Trout, Squeteague
886.91K
9.76M
1.40M
Scup
876.60K
8.77M
2.09M
Mid-Atlantic Bight - 1980
Menhaden, Atlantic
30.78M
646.41M
10.94M
Flounder, Summer
1.15M
83.83M
72.00M
Scup
550.89K
37.46M
15.53M
Mid-Atlantic Bight - 1990
Menhaden, Atlantic
115.86M
4.63B
286.14M
Mackerel, Atlantic
1.67M
103.62M
13.87M
Croaker, Atlantic
1.35M
71.65M
22.53M
Mid-Atlantic Bight - 2000
Menhaden, Atlantic
69.60M
2.64B
167.17M
Croaker, Atlantic
2.16M
106.02M
42.93M
Mackerel, Atlantic
1.70M
59.41M
6.38M
Mid-Atlantic Bight - 2010
Menhaden, Atlantic
118.29M
1.89B
154.46M
Bass, Striped
1.70M
25.56M
75.05M
Croaker, Atlantic
1.08M
24.81M
21.37M
Landings data obtained from the Greater Atlantic Regional Fishing Office (GARFO)
Blackburn, T. M., Gaston, K. J., and Loder, N. 1999. Geographic gradients in body size: a clarification of Bergmann’s rule. BIODIVERSITY RESEARCH. Diversity <html_ent glyph="@amp;" ascii="&"/> Distributions, 5: 165–174. http://dx.doi.org/10.1046/j.1472-4642.1999.00046.x.
Brown, J., West, G. B., and Enquist, B. 2000. Scaling in biology: Patterns and processes, causes and consequences. In pp. 1–24.
Frederikse, T., Landerer, F., Caron, L., Adhikari, S., Parkes, D., Humphrey, V. W., Dangendorf, S., et al. 2020. The causes of sea-level rise since 1900. Nature, 584: 393–397. http://dx.doi.org/10.1038/s41586-020-2591-3.
Glantz, M. H., Katz, R. W., and Nicholls, N. (Eds). 1991. Teleconnections linking worldwide climate anomalies: scientific basis and societal impact. Cambridge University Press, Cambridge [England] ; New York.
Hillaert, J., Hovestadt, T., Vandegehuchte, M. L., and Bonte, D. 2018. Size-dependent movement explains why bigger is better in fragmented landscapes. Ecology and Evolution, 8: 10754–10767. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6262741/.
Kerr, S. R., and Dickie, L. M. 2001. The Biomass Spectrum: A Predator-Prey Theory of Aquatic Production. Columbia University Press.
Meyssignac, B., Boyer, T., Zhao, Z., Hakuba, M. Z., Landerer, F. W., Stammer, D., Köhl, A., et al. 2019. Measuring global ocean heat content to estimate the earth energy imbalance. Frontiers in Marine Science, 6. http://dx.doi.org/10.3389/fmars.2019.00432.
Neto, A. G., Langan, J. A., and Palter, J. B. 2021. Changes in the Gulf Stream preceded rapid warming of the Northwest Atlantic Shelf. Communications Earth & Environment, 2: 1–10. https://www.nature.com/articles/s43247-021-00143-5.
Nye, J., Link, J., Hare, J., and Overholtz, W. 2009. Changing spatial distribution of fish stocks in relation to climate and population size on the Northeast United States continental shelf. Marine Ecology Progress Series, 393: 111–129. http://www.int-res.com/abstracts/meps/v393/p111-129/.
Pershing, A. J., Alexander, M. A., Hernandez, C. M., Kerr, L. A., Le Bris, A., Mills, K. E., Nye, J. A., et al. 2015. Slow adaptation in the face of rapid warming leads to collapse of the gulf of maine cod fishery. Science, 350: 809–812. https://www.science.org/doi/full/10.1126/science.aac9819.
Shin, Y.-J., Rochet, M.-J., Jennings, S., Field, J. G., and Gislason, H. 2005. Using size-based indicators to evaluate the ecosystem effects of fishing. ICES Journal of Marine Science, 62: 384–396. https://doi.org/10.1016/j.icesjms.2005.01.004.
Sprules, W. G., and Barth, L. E. 2016. Surfing the biomass size spectrum: some remarks on history, theory, and application. Canadian Journal of Fisheries and Aquatic Sciences, 73: 477–495. http://www.nrcresearchpress.com/doi/10.1139/cjfas-2015-0115.
---title: "Community Size-Structure of the Northwest Atlantic Groundfish Communities, Response to Direct Disturbance and a Changing Environment"author: name: "Adam A. Kemberling" title: "Quantitative Research Associate" url: https://github.com/adamkemberling affiliation: Gulf of Maine Research Institutedescription: | Size Spectra Analysis of the Northeast US Finfish Communitydate: "`r Sys.Date()`"# format: docxformat: html: self-contained: true code-fold: true code-tools: true df-print: kable toc: true toc-depth: 2editor: sourceexecute: echo: false warning: false message: false fig.height: 6 fig.width: 7 fig.align: center comment: ""bibliography: references.bibcsl: ices-journal-of-marine-science.csl---```{r}#| label: load packages and functions#### Packages ####library(car)library(targets)library(rnaturalearth)library(here)library(sf)library(gmRi)library(patchwork)library(gt)library(knitr)library(tidyverse)library(ggstream)library(broom)library(bcp)library(scales)library(corrplot)library(readxl)library(Hmisc)library(MuMIn)library(gtsummary)library(broom.helpers)# changepoint packagelibrary(EnvCpt)# Package Conflictsconflicted::conflict_prefer("select", "dplyr")conflicted::conflict_prefer("filter", "dplyr")# Support functionssource(here("R/support/sizeSpectra_support.R"))# Resource Pathres_path <- gmRi::cs_path("res")# GGplot themetheme_set(theme_gmri() +theme(panel.grid.major =element_line(linewidth =0.5, linetype =1, color ="gray"),panel.grid.major.x =element_line(linewidth =0.5, linetype =3, color ="gray"),panel.grid.minor =element_line(linewidth =0.5, linetype =3, color ="gray90"),axis.line =element_line(color ="black"),axis.line.y =element_line(color ="black"), strip.background =element_rect(colour =gmri_cols("teal")),legend.position ="bottom"))# Map polygonsus_poly <-ne_states("united states of america", returnclass ="sf")canada <-ne_states("canada", returnclass ="sf")# levels for faceting areasarea_levels <-c("Northeast Shelf", "GoM", "GB", "SNE", "MAB")area_levels_long <-c("Northeast Shelf", "Gulf of Maine", "Georges Bank", "Southern New England", "Mid-Atlantic Bight")# table to join for swapping shorthand for long-hand namesarea_df <-data.frame(area =c("Scotian Shelf", "Gulf of Maine", "Georges Bank", "Southern New England", "Mid-Atlantic Bight", "All"),survey_area =c("SS", "GoM", "GB", "SNE", "MAB", "Northeast Shelf"),area_titles =c("Scotian Shelf", "Gulf of Maine", "Georges Bank", "Southern New England", "Mid-Atlantic Bight", "Northeast Shelf"))# Stupid shapefiles# Get the paths to the shapefiles used to masktrawl_paths <- gmRi::get_timeseries_paths(region_group ="nmfs_trawl_regions", box_location ="cloudstorage")# Polygons for each regionall_poly <-read_sf(trawl_paths[["inuse_strata"]][["shape_path"]]) # Shapefiles for the fisheries stat zonesstat_zones <-read_sf(str_c(res_path, "Shapefiles/Statistical_Areas/Statistical_Areas_2010_withNames.shp"))# Make a list of zones to roughly match the survey areas:fish_zones <-list("Gulf of Maine"=c(511:515, 464, 465),"Georges Bank"=c(521, 522, 525, 561, 562),"Southern New England"=c(611, 612, 613, 616, 526, 537, 538, 539),"Mid-Atlantic Bight"=c(614:615, 621, 622, 625, 626, 631, 632))``````{r}#| label: functions-cell# Function to process summaries for various factor combinationsget_group_summaries <-function(data_in, ...){# Do some grouping to get totals group_totals <- data_in %>%group_by(...) %>%summarise(total_survey_catch =sum(numlen, na.rm = T),total_lw_bio =sum(sum_weight_kg, na.rm = T),total_strat_abund =sum(strat_total_abund_s, na.rm = T),total_strat_lw_bio =sum(strat_total_lwbio_s, na.rm = T), .groups ="drop") # length bins group_lengths <- data_in %>%group_by(..., length_bin) %>%summarise(lenbin_survey_catch =sum(numlen),lenbin_lw_bio =sum(sum_weight_kg),lenbin_strat_abund =sum(strat_total_abund_s),lenbin_strat_lw_bio =sum(strat_total_lwbio_s), .groups ="drop") %>%left_join(group_totals) %>%mutate(perc_total_catch = (lenbin_survey_catch - total_survey_catch) *100,perc_lw_bio = (lenbin_lw_bio - total_lw_bio) *100,perc_strat_abund = (lenbin_strat_abund - total_strat_abund) *100,perc_strat_lw_bio = (lenbin_strat_lw_bio - total_strat_lw_bio) *100)# weight bins group_weights <- data_in %>%group_by(..., weight_bin) %>%summarise(wtbin_survey_catch =sum(numlen),wtbin_lw_bio =sum(sum_weight_kg),wtbin_strat_abund =sum(strat_total_abund_s),wtbin_strat_lw_bio =sum(strat_total_lwbio_s),.groups ="drop") %>%left_join(group_totals) %>% gmutate(perc_total_catch = (wtbin_survey_catch / total_survey_catch) *100,perc_lw_bio = (wtbin_lw_bio / total_lw_bio) *100,perc_strat_abund = (wtbin_strat_abund / total_strat_abund) *100,perc_strat_lw_bio = (wtbin_strat_lw_bio / total_strat_lw_bio) *100)return(list("length_bins"=drop_na(group_lengths),"weight_bins"=drop_na(group_weights)))}# Function to process summaries for various factor combinationslog2_bin_summaries <-function(data_in, ...){# Do some grouping to get totals group_totals <- data_in %>%group_by(...) %>%summarise(total_survey_catch =sum(numlen, na.rm = T),total_lw_bio =sum(sum_weight_kg, na.rm = T),total_strat_abund =sum(strat_total_abund_s, na.rm = T),total_strat_lw_bio =sum(strat_total_lwbio_s, na.rm = T), .groups ="drop") # weight bins group_weights <- data_in %>%mutate(weight_bin = bin_label) %>%group_by(..., left_lim) %>%summarise(wtbin_survey_catch =sum(numlen),wtbin_lw_bio =sum(sum_weight_kg),wtbin_strat_abund =sum(strat_total_abund_s),wtbin_strat_lw_bio =sum(strat_total_lwbio_s),.groups ="drop") %>%left_join(group_totals) %>%mutate(perc_total_catch = (wtbin_survey_catch / total_survey_catch) *100,perc_lw_bio = (wtbin_lw_bio / total_lw_bio) *100,perc_strat_abund = (wtbin_strat_abund / total_strat_abund) *100,perc_strat_lw_bio = (wtbin_strat_lw_bio / total_strat_lw_bio) *100)return(list("weight_bins"=drop_na(group_weights)))}# Function to grab the correlation data and lag dataget_ccf_vector <-function(x,y){# Run the ccf ccf_data <-ccf(x,y,plot= F , lag.max =15)# Get the signif:# https://www.squaregoldfish.co.uk/programming/r_acf_significance.md/# Not 100% sure n is the same for ccf as it is for acf, but... significance_level <-qnorm((1+0.95)/2)/sqrt(sum(!is.na(x)))data.frame("acf"= ccf_data$acf,"lag"= ccf_data$lag,"sigpos"= significance_level,"signeg"= significance_level*-1 )}# Plotting the correlation matrix over a specific decadeplot_period_correlations <-function(x, label) { x %>%filter(response_short !="Spectra Slope", response_area !="All") %>%ggplot(aes(var1, response_short, fill = r2)) +geom_tile() +geom_text(aes(label = signif)) +facet_grid(response_area ~ driver_type, scales ="free_x", space ="free") +scale_fill_distiller(palette ="RdBu", limits =c(-1, 1), breaks =c(-1,0,1)) +coord_cartesian(clip ="off") +theme(axis.text.x =element_text(angle =45, hjust =1, vjust =1),axis.text.y =element_blank(),strip.text.y =element_text(angle =0),legend.position =c(1.205, -.25),legend.title =element_text(vjust =0.75),plot.caption =element_text(hjust =0)) +guides(fill =guide_colorbar(title.position ="top", title.hjust =0.5)) +labs(x ="Correlated Feature", y ="Size Spectrum Slope",fill ="Correlation Coefficient",title =str_c(label, " | Correlation of Size Spectra to Hypothesized Drivers"),subtitle ="Correlation coefficients shown display same-year correlations.",caption =md("* Symbol signals a significant correlation at an alpha = 0.05"))}```### Potential Journals:1. **ICES Journal of Marine Science**# AbstractThe surface waters of the Northwest Atlantic Ocean are among the fastest warming on Earth. This area is highly-productive biologically, and there are concerns that ecological consequences will follow this rapid-warming. Research on the impacts of this rapid warming has primarily focused on high-profile and/or upper trophic level species. Ecological theory and laboratory studies suggest that elevated temperatures facilitate early maturation and smaller adult body-sizes. However, it is unclear whether that relationship might be mitigated against through adaptive behaviors in an open ocean environment. Here we've investigated ecosystem wide impacts on the individual size distribution (ISD) to track changes in community size structure. In cases where community responses are not adequate to counter the impacts of elevated temperatures, we anticipated a steepening of the size spectrum slope (ISD exponent). A steeper relationship relating to a reduction in larger sized individuals and an increased prevalence of smaller sized individuals. Using data from fisheries independent surveys we calculated the community size spectra for four regions along the US NE continental shelf. Correlation/regression analyses were then performed to then assess the degree to which these changes were in alignment to hypothesized bottom-up and top-down disturbances. At the regional scale, we found that community size structure changes (spectra slope) were the largest in the Northern regions, in the Gulf of Maine and Georges Bank. These areas are home to the coldest temperatures and the largest proportions of groundfish species in the community. Spectrum slope declines were most pronounced in the 80's and 90's, before the rapid warming of the last decade. The timing of these declines suggest that external factors drove the initial declines of larger-sized individuals within the communities, before elevated temperatures began to influence the ecosystem. Correlation analyses reveal that while fisheries landings are strongly correlated with these declines, bottom-up factors of zooplankton community metrics, Gulf Stream Index, and SST anomalies are also important. While the primary pressure of fisheries exploitation has declined dramatically over time, the recovery of larger-sized individuals has not been seen. That kind of recovery will likely depend on the elevated temperatures seen over the last decade.# IntroductionUnderstanding the impacts of climate change on local and regional scales is a high priority area of study among the international scientific community. The largest determinant of annual climatic timescale variation is the heat exchange between the atmosphere and the ocean @Glantz1991. Major atmospheric and oceanographic climate modes like the El Nino Southern Oscillation (ENSO), the Gulf Stream Index (GSI) or the North Atlantic Oscillation (NAO) can be linked to localized climate impacts via energy transport between the ocean and atmosphere in a linkage referred to as a "teleconnection" @liu2007. The existence of these pathways demands that scientists understand the inter-connected relationship between local weather and large-scale oceanographic regimes. Scientists following climate change through study of Earth's energy imbalance have observed that over 90% of that excess heat from climate change is accumulating in the ocean @meyssignac2019. This fact has re-focused attention on concerns for the future health of marine ecosystems as temperatures, acidification, sea level, and ocean circulation patterns change (@Johnson2020; @findlay2021; @meyer-gutbrod2021; @frederikse2020; @neto2021). Focusing more narrowly on the impacts of rapid temperature change, scientists have predicted that ecological communities will respond in a number of predictable pathways, each to the limit that their constituent species' life histories can support a response. These responses include shifts in geographic distribution annd alterations to their growth and maturation (Bergmanns Rule, @blackburn1999, **CITE early maturation when warm, @atkinson1995** ) . Marine ecosystems are also shaped directly through human interactions like fishing, which has also been shown to impact these single-species characteristics ( CITE: genetic impacts of size-selective gear, allee effects).Marine ecosystems are complex and are difficult to sample accurately. Much research on the early impacts of climate change on marine systems has focused on individual species or small communities that can be consistently sampled and/or have their interactions explicitly described ( @pinsky2013, @nye2009, @pershing2015 ). An obvious weakness of these approaches is the assumption that all ecological needs provided through inter-species interactions will be equally met under projected climate change conditions. Another lens with which to study the organization of a community is through the study of traits. Size is commonly referenced as the master trait, with well documented relationships with metabolism and bio-energetics @sheldon1972 @brown2000 @white2007 . Marine communities are very-often size-structured, with size correlating to trophic position, making them ideal for study through this lens. Size spectra theory is an ecological framework for studying the relationship between the abundance and size of individuals in a community . This relationship relates how efficiently energy flows from the smallest individuals at the lowest trophic levels to larger individuals occupying higher trophic levels @andersen2006 @hillaert2018 . It is a taxon-agnostic method for detecting changes to ecosystem health and has shown to be sensitive to both large-scale environmental disturbance as well as direct human interaction @shin2005 @kerr2001 . When organized on a log scale, this relationship between abundance and size reveals a linear decline in abundance with increasing body size @white2008 . Highly productive and diverse marine ecosystems are associated with less steep declines in abundance with size, and support larger numbers of higher trophic level individuals. By studying how this relationship changes through time scientists can observe the energy-transfer efficiency of a community as a signal of ecological health and robustness to external changes in the environment @sprules2016 .Our study is focused on changes in the size distribution finfish community of the Northeast US continental shelf. This is a region that is experiencing rapid increases to temperature tied to changes in the behaviors of the Gulf Stream and Labrador Currents @meyer-gutbrod2021 . This ecosystem has been sampled extensively as part of national fisheries management efforts and has long-term records on the size distribution of the finish community, providing a long-history with which to study changes in size structure. Based on early observational studies and ecological theory we anticipate that individual level responses to a rapidly warming environment will be detectable in the size-structure of the region's ecological community, evidence that ecosystems are experiencing change-of-function as a result of rapid environmental change.## Rapid Warming of Northeast ShelfThe Northwest Atlantic is one of the fastest warming locations in the global oceans. Sea surface temperatures in the Gulf of Maine since 1982 have been warming at rates faster than 96% of the world's oceans, with similar warming rates along the northwest Atlantic shelf (Pershing et al. (2018)). The persistent elevated temperature regime of the area is a result of several forces: a combination of shifting ocean currents and the unique bathymetry of the region. A Northward shift in the Gulf Stream directly increased the regional temperatures through increased transport of warm Gulf Stream water into areas like the Gulf of Maine. The Northward Gulf Stream shift is associated with a higher frequency of warm core rings, and the obstruction of cold-water Scotian Shelf current flow that otherwise counters the influence of the Gulf Stream on the region's temperatures (Gangopadhyay et al. (2019); Meyer-Gutbrod et al. (2021)). The combination of these oceanographic changes has led to a warmer continental shelf habitat, which has caused alarm among scientists studying the ecology of the region.Temperature plays a critical role on biological function, affecting many of the chemical reactions that underpin basic physiological function. Reactions impacting critical actions like locomotion and feeding behavior, metabolism and development, and even maturation rates. Because of these relationships, species have evolved thermal preferences around which these functions each operate efficiently. Individuals that are unable to maintain their thermal preferences internally risk metabolic costs unless they can follow their thermal preference in the environment through locomotion or adapt to less-favorable conditions through changes in behavior. In an era of anthropogenic climate change, there is an expectation that many species will be displaced from historic habitats in their efforts to follow their thermal preferences. Recent research in marine environments has shown evidence of this, as species are now shifting to higher latitudes and to deeper depths in the pursuit of more favorable conditions (Nye et al. (2009); Pinsky et al. (2013)). Other work suggests that temperature related impacts may manifest through physiological changes and changes in seasonal phenology, hindering species conservation efforts (Miller et al. (2018); Pershing et al. (2015); Meyer-Gutbrod et al. (2021)). Need to connect temperature to size here: Direct quote from Guiet et al. (2016) , but nails the connection back to temp expectations:> Because it controls chemical reactions, temperature controls metabolic rates which underpin maintenance, growth or reproduction (Clarke and Johnston, 1999; Kooijman, 2010) as well as the functional responses to food density (Rall et al., 2012). Guiet et al. (2016)... In addition to the impact of temperature on communities' intercepts (heights), the impact of temperature on the speed of the energy flow within communities may affect other properties, such as their resilience to perturbations or the intensity of trophic cascades (Andersen and Pedersen, 2009).The potential for elevated temperatures to impact the size structure of an ecosystem has implications for the ecosystem resilience in the face of climate change, as well as the blue economies & natural resource systems that rely upon their good health.## Size Spectra TheorySize is a defining characteristic of species and mediates many ecological interactions and metabolic pathways (Brown et al. 2000) . Size is a major determining factor for the mobility of an organism which directly impacts the ability to evade predation, find foraging success, and efforts to locate and follow seasonal habitats. Size and physical dimensions also impact the metabolic costs associated with each of these behaviors (Hillaert et al. 2018), mediating exchange rates with the immediate environment like heat loss or desiccation in terrestrial species (Gillooly et al. (2001); Heatwole et al. 1969). Body size even informs life history features like life span and the trophic position an individual might occupy through its impact on metabolism and resource use (White et al. 2007). Size structured environments are a fundamental organizational pattern that emerges from these relationships add_citation. Within strongly size-structured ecosystems, growth and maturity changes alter fitness and ultimately determine whether a species is successful in that environment add_citation . Ecological theory is rich with models relating how energy transfers from smaller prey species to larger predatory trophic levels, the allocation of energy for growth, and the trade offs of allocating energy towards those ends (Bertalanffy (1938); Bertalanffy (1957)); add_size_theory_citations). A globally persistent pattern in ecology entangled in those relationships and their critiques is the decline in abundance with increasing body size (Damuth (1981); Currie (1993); Sheldon et al. (1972); Loeuille and Loreau (2006)). The relationship between size and abundance integrates processes operating on the cellular, individual, and community levels simultaneously. By measuring how this relationship changes, scientists can detect whether the fundamental energy transfer pathways from small individuals to larger predators have been impacted or altered, an outward indication of ecosystem-wide changes. The quantities for size and abundance are often the most readily collected data assets of any ecological community. This creates an opportunity to learn much about a system from a relatively low-effort in data collection. For these reasons, size spectrum analyses and individual size distribution (ISD) methods have gained increasing attention as an entry point to assessing ecosystem health and to detect system-wide disturbance (Shin et al. (2005)'; Pomeranz et al. (2022); White et al. (2007)). Spectra models are also taxon agnostic and avoid the need to explicitly articulate each predator-prey interaction. The "size spectrum" describes the distribution of abundance or biomass as a function of individuals' mass on a log--log scale (Guiet et al. (2016); Kerr and Dickie (2001)) . Size spectra condense the complexities of predator prey networks and their interactions into taxon-agnostic size-based indices. These indices capture the emergent properties of a system, and have become increasingly used as indicators of ecosystem health. Within the context of fisheries management, changes in spectrum slopes have been associated with fishing exploitation, primarily through the targeted removal of larger individuals (Bianchi et al. (2000); Shin et al. (2005)). Numerical experiments have also linked changes in slope to environmental disturbances (Guiet et al. (2016)). Size spectra have also been shown to express predictable relationships between ecosystems of similar productivity levels as well as from distinct temperature regimes (Guiet et al. (2016)). Use Sprules and Barth Paper to discuss applications Use Pomerantz paper & Edwards to extend into ISD## External Drivers of Size StructureOur study area's ecology is one that has historically been studied through the fishing grounds it once supported and their challenges. The focus has traditionally been on the success/failure of single species management actions, however there are growing efforts to better understand the full ecological impact of this endeavor. Fishing practices are often size-selective, with harvesters targeting larger individuals for a higher yield on any time and capital invested. Larger individuals are also commonly the older individuals in a population, having lived longer to achieve those sizes. With the unfortunate side-effect of having an outsized impact on the reproductive potential of the population. Larger individuals have a greater impact on population resilience and recovery, capable of holding more (and often of higher quality) eggs which increase the odds of recruitment success. Heavily fished populations have been shown to exhibit changes to growth rates, suggesting an additional genetic impact from size-selective fishing on the target-species. The direct removal of larger individuals carries a predictable impact on the size distributions of the target species, but may also change that of non-target species as well. Non-target species may be impacted as by-catch from the fishery, or may alternatively find relief from predation if the target species is a direct predator. This can even create circumstances where prey species can crowd out, or outcompete the juveniles of their natural predators, reinforcing the new community. Industrial fishing practices are inherently size-selective, with an outsized footprint relative to small-scale fisheries. Large removals have an immediate and measurable impact on the community size-distribution with potential additional impacts on the future population as well. Size-based harvest in fisheries has been shown to create selective pressures that promote characteristics of early maturation at smaller sizes add_citation. These consequences may independently or in combination create circumstances where a community is unable to return to conditions before these disturbances occurred. reference gb/ss spectra early work (Duplisea & Kerr 1995, or Kerr book)The ecological community in our study area has with no uncertainty been altered through human behaviors over the last century. Early research estimated that biomass had more than halved in some areas by the 1960's, pre-dating federal monitoring efforts and inspiring their formation (Fogarty and Murawski (1998)). Key stocks that supported international fishing efforts collapsed, and recovery efforts have in many species fallen short. Work exploring retrospective patterns in stock assessment predictions have highlighted that these shortfalls can be partially attributed to un-accounted for influences of a changing environment. However, other research has noted that the community structure itself may have changed, and with it, how it responds to outside pressure. That same early research noted that species replacement of commercial target species by skate and dogfish was happening as their numbers quickly began to rise. A wealth of anecdotal evidence and records from commercial landings support the conclusion that the region does not support either the fisheries or the fishing communities in the same way today. The Lobster industry remains King in the Gulf of Maine, another side-effect of the decline in large predators in the area. With other similar declines in large-fish populations up and down the coast, scientists have recently brought the focus of their attention away from single-species management to more holistic and ecosystem-wide approaches.## Regime Shifts in Marine Systems**Goals:**Want to hit: dynamic tipping points, hysteresis, and non-stationary functional relationships. Show examples of the idea on single species, suggest that we size spectra can be used to check the whole community for tipping pointsWhen an ecosystem experiences an abrupt change in its internal dynamics in response to internal or external pressures, it is said to have crossed a tipping point. When these tipping points are crossed, it often becomes difficult or impossible to return to the previous state. This is true even when the pressures that caused the shift have ceded and favorable conditions have returned, a phenomenon known as hysteresis. This phenomenon has been observed in individual population trends, and is a major factor undermining sustainable management of fisheries (Blocker 2023; Sguottti 2019) . Another feature that may accompany a regime shift is the altering of functional relationships with other biotic or abiotic features of the environment. When this occurs, the productivity of a species and/or its resilience to stress may be altered, and assumptions underpinning sustainable harvest thresholds should be updated (Blocker 2023). At the community level the alternative community states may support a different composition of species. The community itself may respond differently to the same stressors/stimulus. In the Northeast Shelf Region there is evidence to suggest that through a combination of human influences and environmental changes that the ecology of some areas may have crossed ecological tipping points. This study explores whether evidence for an ecological tipping point can be seen through the perspective of the size spectrum, and whether there is evidence for hysteresis, or non-stationarity in the community's relationships to external forces.### PurposeUnderstanding that human populations depend on the health of their ecosystems, there is a need to better understand when we've crossed ecological tipping points. One way to capture these changes is through the use of community-wide metrics that capture the inner-efficiencies of many trophic interactions. Size based indices are metrics that can be estimated from the information that has historically been available from long-term survey efforts. These indices have been shown to be sensitive to the impacts of fishing, but also capture environmental dynamics as well. We estimated size spectrum relationships to use as size based indicators of the ecological communities within each sub-region of the Northeast US continental shelf. Based on ecological theory we believe that the sustained increases in temperature in the NW Atlantics should have a physiological impact on the community size structure. We hypothesize that rapid warming alters the community through the direct influence of temperature on metabolism, growth, and population productivity. We also hypothesize that size spectra will uniquely reflect the physical environment of the regions, with less steep size spectra in regions with colder temperatures and of higher primary productivity. Understanding that temperature is not the sole mechanism for altering community size structure, we will explore how the temporal structures in size spectra have changed in relation to major external drivers. We posit that distinct patterns of deviation from that signature may reflect a fundamental change in the ecology of that region, an indication of a possible ecological regime change, and an altered relationship to human and environmental disturbances.# Methods## Fish Data Source and Processing```{r}#| label: load-survdat# 1. Biological data used as inputwithr::with_dir(rprojroot::find_root('_targets.R'), tar_load(catch_log2_labelled)) # rename and format# Add the area titlescatch_size_bins <- catch_log2_labelled %>%fill_func_groups() %>%left_join(area_df) # Get total number of speciesn_species <-length(unique(catch_size_bins$comname))```Data on the biomass, abundance, and size of fish on the Northeast U.S. Shelf were collected as part of the Northeast Fisheries Science Center's bottom trawl survey (Grosslein 1969, Azarovits 1981, Politis 2014). This survey is conducted from Cape Hatteras, North Carolina to the Gulf of Maine each year in the spring and in the fall. The survey follows a stratified random sampling design, with strata defined based on depth, bottom habitat, and latitude. Trawls are performed for a fixed duration at each station, reporting aggregate abundance and biomass for all species caught, and measuring individual lengths and weights for the catch of each species or a sub-sample if that catch is large. Correction factors were applied to aggregate species abundance and biomass to account for changes in vessels, gear, and doors used in the survey over time (Sissenwine and Bowman 1978, Byrne and Forrester 1991, Miller et al. 2010). However, abundance and biomass at length needed to be estimated after these aggregate corrections. As such, abundance at length for each species was adjusted to match the corrected aggregate species abundance at each station, such that for each species, the sum of the resulting estimated abundance numbers across each length is equal to the corrected aggregate abundance.```{r}#| label: study-area-maps#| eval: false#------------- Trawl Strata map --------------# Make the Mapstrata_map <-ggplot() +geom_sf(data = us_poly) +geom_sf(data = canada) +geom_sf(data = all_poly, aes(fill = survey_area), alpha =0.8) +coord_sf(xlim =c(-76.4, -64.4), ylim =c(35, 45.5), expand = F) +scale_fill_gmri() +theme_bw() +map_theme(legend.position =c(0.6, 0.125), legend.background =element_rect(fill ="white"),plot.caption =element_text(hjust =0),plot.margin =margin(t =0, b =0)) +guides(fill =guide_legend(title ="", nrow =2)) # ----------- Landings Statistical Zones -----------------# Trim out what we don't need and label themstat_zones <- stat_zones %>%mutate(survey_area =case_when( Id %in% fish_zones$"Gulf of Maine"~"Gulf of Maine", Id %in% fish_zones$"Georges Bank"~"Georges Bank", Id %in% fish_zones$"Southern New England"~"Southern New England", Id %in% fish_zones$"Mid-Atlantic Bight"~"Mid-Atlantic Bight")) %>%filter(survey_area %in%c("Georges Bank", "Gulf of Maine", "Southern New England", "Mid-Atlantic Bight"))# Map it out? - takes forever from Chesapeake baygarfo_map <-ggplot(stat_zones) +geom_sf(aes(fill = survey_area), alpha =0.8) +geom_sf(data = us_poly) +geom_sf(data = canada) +coord_sf(xlim =c(-76.4, -64.4), ylim =c(35, 45.5), expand = F) +scale_fill_gmri() +theme_bw() +map_theme(legend.position =c(0.6, 0.125),legend.background =element_rect(fill ="white"),plot.caption =element_text(hjust =0),plot.margin =margin(t =0, b =0)) +guides(fill =guide_legend(title ="", nrow =2)) # ggsave(garfo_map, filename = here::here("R/nmfs_size_spectra/images/garfo_regions.png"), bg = "white")# Map them togetherdual_map <- (strata_map | garfo_map) +plot_layout(guides ="collect")# # Save the map:# ggsave(dual_map,# filename = here::here("R/nmfs_size_spectra/images/dual_region_map.png"),# bg = "white",# width = unit(6, "in"),# height = unit(5, "in"), dpi = "retina")```{fig-align="center"}### Community Composition & Functional GroupsAnalyses were performed using 68 species. These species were selected based on the availability of published weight-at-length relationships (Wigley et al. 2003) and represented 98.98% of the total biomass caught in the survey. Each species was assigned to a functional group based on life history and geography using the definitions of (Hare et al. (2016)). Functional groups included coastal fishes, diadromous fishes, elasmobranchs, groundfish, pelagic fishes, and reef fish species (Table 1.). Six species with available length-weight details did not have a functional group designation, these species were designated as reef species. Exploratory analyses showed that the pelagic species biomass was low in all regions, and is unlikely to be representative of true biomass trends due to gear selectivity.```{r}#| label: biomass-group-summaries# This cell provides the group summaries for how much biomass exists in each size/functional group# Actual log10 bins# Do some grouping by year survey area and functional groupfgroup_area <-log2_bin_summaries(data_in = catch_size_bins, Year, survey_area, hare_group)fgroup_all <-log2_bin_summaries(data_in = catch_size_bins, Year, hare_group)fgroup_all <- fgroup_all %>%map(function(x){mutate(x, survey_area ="Northeast Shelf")})# Join the overall values back infgroup_area <-map2(fgroup_area, fgroup_all, bind_rows)# Add new units in millionsfgroup_area$weight_bins <- fgroup_area$weight_bins %>%mutate(strat_abund_mill = wtbin_strat_abund /1e6,strat_lwbio_mill = wtbin_strat_lw_bio /1e6,survey_area =factor(survey_area,levels = area_levels))```Published length-weight relationships ( Wigley et al. 2003 ) were used to convert from length data, available for all individuals, into their corresponding biomass-at-length. To account for differences in sampling effort among survey strata, all corrected abundance-at-length data were area-stratified. Area-stratified biomass-at-length values were then computed as the product of area-stratified abundance-at-length and estimated weight-at-length. All analyses were performed using area-stratified abundances and their associated area-stratified biomass estimates, hereby referred to as simply abundance-at-length & biomass-at-length## Community MetricsOur analyses used all data collected during the spring and fall surveys from 1970-2019. Data were grouped using survey-design strata into four sub-regions: Gulf of Maine, Georges Bank, Southern New England, Mid-Atlantic Bight (Figure 1.). These sub-regions have been widely used in regional ecological studies (e.g., ). For each region, we developed the following time series of ecological indicators:1. Annual mean abundance and biomass by functional group2. Annual mean length and weight of the aggregate community and for each functional group3. Annual estimates of the community size spectrum slope### Quantifying Body Size ChangesChanges in the size structure of the community was measured using the average length and weight across all species, each weighted by their species and size-specific catch rates. The average body length (cm) and body weight (kg) was calculated for each region and within each functional group.$$X_j = \frac{ \sum_{}^{} n_iX_{ij} }{ \sum~w_i }$$ Where $X$ is the body-size metric, $j$ is the year, & $n$ is the abundance for each station $i$.Data for body size trends were not truncated using any minimum or maximum size and reflect all available catch data for the 68 species in this study for which biomass-at-length could be estimated.```{r}#| label: body size data# Load the average body size datawithr::with_dir(rprojroot::find_root('_targets.R'), tar_load(mean_sizes_ss_groups)) # Overall across all regionsshelf_size <- mean_sizes_ss_groups %>%filter(`group ID`=="single years") %>%mutate(survey_area ="Northeast Shelf")# Grouped on year and regionregional_size <- mean_sizes_ss_groups %>%filter(`group ID`=="single years * region") %>%bind_rows(shelf_size) %>%mutate(Year =as.numeric(Year),survey_area =as.character(survey_area)) %>%left_join(area_df) %>%mutate(area_titles =factor(area_titles, levels = area_levels_long),area_copy = survey_area)# 2. Grouped on year, region, & functional groupfunctional_group_size <- mean_sizes_ss_groups %>%filter(`group ID`=="single years * region * functional group") %>%mutate(Year =as.numeric(Year),survey_area =factor(survey_area, levels = area_levels))# Refactor areasfunctional_group_size <- functional_group_size %>%mutate(survey_area =as.character(survey_area)) %>%left_join(area_df) %>%mutate(area_titles =factor(area_titles, levels = area_levels_long),area_copy = survey_area)``````{r}#| label: plotting-bigfish-smallfish#| eval: true# Idea: it would be great top be able to speak more directly to population trends for large/small individuals within a functional group...# One avenue is to break it into large/small# Could use the median body sizesize_trends <- catch_size_bins %>%mutate(fish_size =ifelse(wmin_g >1000, "Over 1kg", "Less than 1 kg")) %>%group_by(Year,area, hare_group, fish_size) %>%summarise(total_abund =sum(strat_total_abund_s)/1e6,total_biomass =sum(strat_total_lwbio_s), .groups ="drop") %>%pivot_longer(cols =c(total_abund, total_biomass), names_to ="var", values_to ="vals") # Abundancesize_trends %>%filter(var =="total_abund") %>%#filter(var == "total_biomass") %>% filter(hare_group %in%c("groundfish", "elasmobranch")) %>%ggplot(aes(Year, vals)) +geom_point(aes(color = fish_size), size =0.5) +geom_smooth(aes(color = fish_size), linewidth =1, se = F) +scale_color_gmri() +scale_y_continuous(labels = scales::comma_format()) +facet_grid(area~hare_group) +labs(color ="Fish Size", y ="Abundance (millions)" )# Biomasssize_trends %>%#filter(var == "total_abund") %>% filter(var =="total_biomass") %>%filter(hare_group %in%c("groundfish", "elasmobranch")) %>%ggplot(aes(Year, vals)) +geom_point(aes(color = fish_size), size =0.5) +geom_smooth(aes(color = fish_size), linewidth =1, se = F) +scale_color_gmri() +scale_y_continuous(labels = scales::comma_format()) +facet_grid(area~hare_group) +labs(color ="Fish Size", y ="Biomass (kg)" )```### Estimating Size Spectra```{r}#| label: load the size spectrum indiceswithr::with_dir(rprojroot::find_root('_targets.R'), tar_load(size_spectrum_indices)) # Grab the overall shelf-wide SSshelf_ss <- size_spectrum_indices %>%filter(`group ID`=="single years") %>%mutate(survey_area ="Northeast Shelf")# Grab SS Groups for each regionregion_indices <- size_spectrum_indices %>%filter(`group ID`=="single years * region") %>%bind_rows(shelf_ss) %>%mutate(yr =as.numeric(as.character(Year)),survey_area =factor(survey_area, levels = area_levels),sig_fit =ifelse(log2_sig_strat <0.05, "Significant", "Non-Significant"))# Make a copy so we can gray out the lines in plots# Merge the long names for labels "area_titles"region_indices <- region_indices %>%mutate(survey_area =as.character(survey_area)) %>%left_join(area_df) %>%mutate(area_titles =factor(area_titles, levels = area_levels_long),area_copy = survey_area)```Abundance size spectra were calculated using the area-weighted abundance at length information from the catch data. Lengths of individuals in the catch data are measured to the nearest cm, with smaller specimens measured to the nearest millimeter. Because individual biomass is estimated from those length measurements, there is a range of possible body mass values between the increments used (cm and mm). The relationship between length and mass in fishes is exponential and taxon specific. Using only the lower or upper end of those ranges introduces biases which are different for each species and increase with larger body-sizes. To account for these biases, we used the extended likelihood method (MLEbin) of Edwards et al. (2020). This method estimates the exponent of size spectra (b) for a bounded power law relationship between abundance and length-estimated biomass. The exponent of this relationship is analogous to slope estimates from linear regression on logarithmic axes, with a steeper slope (i.e. a more negative b) indicating fewer large-bodied individuals and/or more small-bodied fishes (Carvalho 2021) . This method has been shown to be the most accurate for estimating the exponent of size spectra when tested against alternative methods that require arbitrary decisions around binning the data and make estimates difficult to compare directly across studies (White 2008; Sprules & Barth 2016; Edwards 2017).$$\begin{align*}f(x) = \frac{ (\lambda + 1)x^{\lambda} }{ x^{\lambda+1}_{max} - x^{\lambda+1}_{min} }~~~~~~\lambda\neq1,\\\\f(x) = \frac{1}{logx_{max} - logx_{min}}~~~~~~\lambda=1\end{align*}$$ {#eq-ISD}Using this method, the size spectra exponent (b) was estimated for each region from 1970 to 2019. A minimum biomass of 1g was used for the lower bound and a maximum biomass of 10kg was used as an upper bound for the ISD's bounded power law probability density function Equation 1, where X is body mass & Λ is the scaling exponent of the abundance density function. This body-size range constitutes 97.83% of all the biomass for the 68 species included in this study. This body-size range truncation was done to account for poor gear selectivity at the smallest and largest size ranges. This measure also imposes shared bounds to the size range covered by our size spectra, reflecting only the ranges we'd expect to consistently sample across these different areas. Exponents of size spectra (b) were calculated using code modified from the sizeSpectra package (Edwards et al. (2017); Edwards et al. (2020)) and implemented using the R statistical programming language.## Drivers of Size Spectra ChangesThe impact of external factors on the changes in size spectra were explored using multiple regression analyses. Annual variation in size spectrum slopes was modeled using several hypothesized drivers including the environmental state and measures of anthropogenic disturbances. The large-scale environmental drivers included were regional SST anomalies & the larger-scale impact of the Gulf Stream Index (GSI). Fishing pressure represents the primary top-down anthropogenic driver in the region, and was investigated using the aggregate regional commercial landings data. Bottom-up ecological interactions were explored using zooplankton indices from the EcoMon plankton survey. Each of these drivers (climate, productivity, fishing) have independently been shown in other works to have measurable impacts on size spectra. Our exploratory investigations here are an effort to evaluate which best explain the trends within our study region, and whether those relationships have changed over the study period.### Gulf Stream IndexData for the Gulf stream index (GSI) was obtained from the ecodata package in R (Bastille & Hardison 2018) . This package supplies GSI data at monthly intervals following the methodology of Pérez-Hernández and Joyce (2014) and Joyce et al. (2019) , using as sea level height anomaly data from the Copernicus Marine Environment Monitoring Service.```{r}#| label: gsi-data-prep#| eval: true#### 2. Climate Drivers ##### From ecodata: GSI & Zooplankton#### GSI# Gulf Stream Indexgsi <- ecodata::gsi %>%mutate(Time =str_pad(as.character(Time), width =7, side ="right", pad ="0"),year =as.numeric(str_sub(Time, 1, 4)),month =str_sub(Time, -2, -1),Time =as.Date(str_c(year, month, "01", sep ="-")))```### Sea Surface Temperature DataGlobal sea surface temperature (SST) data were obtained via NOAA's Optimum Interpolation SST analysis (OISSTv2), which provides daily SST values at a 0.25° latitude x 0.25° longitude resolution (Reynolds et al. 2007). A daily climatology for every 0.25° pixel in the global data set was created using average daily temperatures spanning the period of 1982-2011. Daily anomalies were then computed as the difference between observed temperatures and the daily climatological average. OISSTv2 data used in these analyses were obtained from the NOAA Physical Sciences Laboratory, Boulder, Colorado, USA from their website at https://psl.noaa.gov/data/gridded/data.noaa.oisst.v2.html.SST data were regionally averaged to match the four survey sub-regions described above (Figure 1a), producing daily time series for each area. These were then averaged into annual time series of surface temperatures and anomalies. All averaging between satellite grid units was done with area-weighting of the grid cells to account for differences in cell size with latitude/longitude in the OISSTv2 data.```{r}#| label: load temperature data# Load the regional temperatures from {targets}withr::with_dir(rprojroot::find_root('_targets.R'), tar_load(regional_oisst))# Get regional averagestemp_regimes <- regional_oisst %>%filter(yr >1981) %>%mutate(regime =ifelse(sst_anom >0, "hot", "cold"),survey_area =ifelse( survey_area =="all", "Northeast Shelf", survey_area),survey_area =factor(survey_area, levels = area_levels)) ```### Commercial FishingFishing pressure in the region was indexed using state and federal commercial fishing landings. These data were obtained from the Greater Atlantic Regional Fisheries Office (GARFO) for statistical areas that are routinely used for fisheries reporting and management (Figure 1., right). Individual statistical areas were aggregated into regions that closely align with the survey areas we defined for the size spectra analyses (Figure 1 b.). Each region's commercial landings timeline was scaled using z-score normalization```{r}#| label: garfo-data-prep#| eval: true#### GARFO Landings ##### Load the GARFO landings data:# Landings of finfish* sheet 5landings <-read_xlsx(path =str_c(res_path, "GARFO_landings/KMills_landings by area 1964-2021_JUN 2022.xlsx"), sheet =5) %>%rename_all(tolower)# Add the labels into the landings data and remove what we don't need there:landings <- landings %>%mutate(survey_area =case_when(`stat area`%in% fish_zones$"Gulf of Maine"~"Gulf of Maine",`stat area`%in% fish_zones$"Georges Bank"~"Georges Bank",`stat area`%in% fish_zones$"Southern New England"~"Southern New England",`stat area`%in% fish_zones$"Mid-Atlantic Bight"~"Mid-Atlantic Bight")) %>%filter(survey_area %in%c("Georges Bank", "Gulf of Maine", "Southern New England", "Mid-Atlantic Bight")) %>%mutate(survey_area =factor(survey_area, area_levels_long))# Get Summarieslandings_summ <- landings %>%drop_na() %>%rename("weight_lb"=`landed lbs`,"live_lb"=`live lbs`) %>%drop_na() %>%group_by(year, survey_area) %>%summarise( across(.cols =c(value, weight_lb, live_lb), .fns =list(mean =~mean(.x , na.rm = T), total =~sum(.x , na.rm = T)), .names ="{.fn}_{.col}"), .groups ="drop") # Scale the landings to create an index by arealandings_summ <- landings_summ %>%group_by(survey_area) %>%mutate(total_wt_z =as.numeric(base::scale(total_weight_lb))) %>%ungroup()```### Zooplankton IndicesZooplankton data was from the National Oceanographic and Atmospheric Administration Marine Resources Monitoring, Assessment and Prediction (MARMAP) program and Ecosystem Monitoring (EcoMon) cruises detailed extensively in Kane (2007), Kane (2011), and Morse et al. (2017). Abundance anomalies were computed from the expected abundance on the day of sample collection. The small copepod index is computed by averaging the individual abundance anomalies of Pseudocalanus spp., Centropages hamatus, Centropages typicus, and Temora longicornis. The "large-copepod" anomaly values are the abundance anomaly of Calanus finmarchicus, the largest copepod in the Northeast U.S. region. Anomalies for both large and small zooplankton groups were averaged within ecological production units by the data provider, producing annual timeseries for the Gulf of Maine, Georges Bank, and Mid-Atlantic Bight. The timeseries for the Mid-Atlantic Bight EPU was used as a predictor for both the Mid-Atlantic Bight and Southern New England regions in our analyses. This decision was made to address how the spatial coverage of this EPU spans the joint area of our two southernmost regions.```{r}#| label: zp-data-prep# ADD Zooplankton: small and large zooplankton anomalies for the EPUS# Don't need to scale# Load original datazp_sli <- ecodata::zoo_sli_anom %>%filter(EPU !="SS")# Reformat to match the other indiceszp_index <- zp_sli %>%#EPUS combine SNE and MAB, we can repeat the values herebind_rows(filter(zp_sli, EPU =="MAB") %>%mutate(EPU ="SNE")) %>%mutate(survey_area =case_when( EPU =="GB"~"Georges Bank", EPU =="GOM"~"Gulf of Maine", EPU =="MAB"~"Mid-Atlantic Bight", EPU =="SNE"~"Southern New England"),Units =str_c("zp_", Var)) %>%select(year = Time, survey_area, Value, Units) %>%pivot_wider(names_from = Units, values_from = Value)``````{r}#| label: combine-drivers-table#### Build a Common form for all Drivers: ##### Format: year, area, var, value#---#### Reshape all the inputs so that we can bind them together in one dataframe# 1.Climate Features# Join the climate modes togetherclim_drivers <- gsi %>%filter(EPU =="All")# Put climate drivers on annual scheduleclim_idx <- clim_drivers %>%group_by(year, area = EPU, var = Var) %>%summarise(value =mean(Value, na.rm = T),.groups ="drop")#---# 2. GARFO landingslandings_idx <- landings_summ %>%group_by(year, area = survey_area) %>%summarise(value =mean(total_wt_z, na.rm = T),.groups ="drop") %>%mutate(var ="landings")#---# 3. Temperaturesst_idx <- temp_regimes %>%left_join(area_df, by ="survey_area") %>%select(year = yr, area, value = sst_anom) %>%mutate(var ="sst_anom")#---# 4. Zooplanktonzp_idx <- zp_index %>%pivot_longer(cols =c(zp_small, zp_large), names_to ="var", values_to ="value") %>%rename(area = survey_area)#---# 4. Size Spectrum Slopes & Interceptsss_idx <- region_indices %>%select(year = Year, survey_area, spectra_slope = log2_slope_strat, #spectra_int = log2_int_strat,isd_slope = b) %>%pivot_longer(# cols = c(spectra_slope, spectra_int, isd_slope), cols =c(spectra_slope, isd_slope), names_to ="var", values_to ="value") %>%mutate(year =as.numeric(year)) %>%left_join(area_df, by ="survey_area") %>%select(-survey_area)#---# All Metrics Togetherall_drivers <-bind_rows(list( clim_idx, landings_idx, sst_idx, zp_idx, ss_idx))``````{r}#| label: driver-scaling# In this chunk the drivers get reshaped to scale# Scaling should cover the period of analysis 1982-2019# SST and GSI (all index/anomaly fields) are not scaled# Put the data in a wide form for scaling:drivers_wide <- all_drivers %>%filter(year >=1982, year <=2019) %>%mutate(id =str_replace_all(str_c(area, "_", var), " ", "_")) %>%select(-c(area, area_titles, var)) %>%pivot_wider(names_from = id, values_from = value) %>%column_to_rownames(var ="year")# Scale the landings, not SST or GSIto_scale <-select(drivers_wide, ends_with("landings")) %>%scale() %>%as.data.frame()dont_scale <-select(drivers_wide, -ends_with("landings"))# Join back togetherdrivers_scaled <-bind_cols(dont_scale, to_scale) %>%rownames_to_column(var ="year")```## Exploring Influence of Drivers on Size SpectraThe hypothesized drivers for size spectrum changes were evaluated using multiple regression analysis. Candidate models for each region were informed by the any observed temporal structures in the size spectra changes through time, and with the inclusion of important lag-effects of predictors.These modeling decisions were informed by the following steps performed prior to multiple regression analysis with the above drivers.### Temporal Lags in Hypothesized DriversIn addition to exploring their contemporaneous impacts, the importance of any time-lags on predictors was also evaluated. Important lagged relationships were identified through the use of cross correlation function (CCF) estimates. These estimates looked at correlations at one-year lag intervals to investigate potential time-delayed effects between the hypothesized drivers and the community's size spectra. CCF were performed between the dependent variables of the size spectra slope with the independent variables of the annual gulf stream index, the corresponding regional commercial landings and sea surface temperature anomalies, and both small and large zooplankton indices. This was performed for annual lags up to 10 years. Significant lags identified in the CCF analysis were then included as additional candidate predictors in the multiple regression analysis of spectrum slope changes.### Non-Stationary Behavior Across Response RegimesTo highlight potential signatures of any non-stationary or non-linear discontinuous dynamics, a hallmark of regime-shift dynamics (Blocker 2023), we explored discontinuous changes in the size spectra using breakpoint analyses. Breakpoint analysis was performed on the annual time series of spectrum slopes to explore their temporal structures for evidence of any potential regime shifts. Breakpoint analyses were performed using the envcpt package (Killick et al. 2021). This package applies an automatic model selection process testing support for candidate model structures including constant/piecewise changes to the mean, variance, trends, & autocorrelations and any identified changepoints located using the pruned exact linear time algorithm (Rikardsen 2004). Best supported model structures from this procedure were then used to inform whether (if any) regional spectra expressed distinct periods of behavior. In the events where changepoints were identified, this was used as evidence that there may have been a community regime change. Following this thinking it is plausible for there to be inconsistencies or non-stationarity in the predictors' impact on the community spectra, as different community regimes may be differently vulnerable to top-down pressures like commercial fishing. To allow for changes in these relationships across different regimes, an additional interaction term was added to the regression model candidates for that region.### Regime Specific CorrelationsBased on any breakpoints detected, single driver correlations were prepared to visually explore how each driver was correlated within the different regimes. Additionally, spectra slope changes were also correlated with the body-size changes of each functional group as a window into what body-size changes might be informing the change in the size structure.```{r}#| label: driver-correlation-periods# Pull different years into a listdriver_periods <-list("1982-1998"=c(1982:1998),"1999-2006"=c(1999:2006),"2007-2019"=c(2007:2019),"1982-2019"=c(1982:2019))# -------------------------------------------------# Idea: Add functional_group_sizes into the correlation chart:# Reshape the size information so we can combine it with the drivers:# Values are scaled heresizes_wide <- functional_group_size %>%select(year = Year, survey_area, hare_group, avg_weight = mean_wt_kg) %>%pivot_longer(names_to ="var", values_to ="value", cols =c(avg_weight)) %>%inner_join(area_df, by ="survey_area") %>%select(-survey_area) %>%mutate(id =str_replace_all(str_c(area, "_", hare_group, "_", var), " ", "_")) %>%select(-c(area, area_titles, var, hare_group)) %>%pivot_wider(names_from = id, values_from = value, values_fn = {mean}) %>%column_to_rownames(var ="year") %>%scale() %>%as.data.frame()# Join regional size changes to the scaled driversdrivers_and_sizes <- drivers_scaled %>%left_join(rownames_to_column(sizes_wide, var ="year"), by ="year") %>%column_to_rownames(var ="year")# Split them out into a list that contains data for each period# Takes the years in driver_periods and pulls the rows that corresponddriver_period_list <-map(driver_periods, function(yrs){ period_matrix <- drivers_and_sizes[rownames(drivers_and_sizes) %in% yrs, ] %>%drop_na()})```### Multiple Regression Modeling of Hypothesized DriversFollowing these explorations into temporal structures, multiple regression models were developed and evaluated independently for each region. Potential drivers included in the models were the annual Gulf Stream Index, the annual regionally averaged SST anomaly value, regionally averaged small and large copepod indices, and the scaled annual commercial landings from that region. Gulf stream index, zooplankton indices, and SST anomalies were not standardized prior to use in model fitting and evaluation as the two former are already an index value and the latter is scaled to a thirty-year climatological reference period. In the event that CCF analysis identified an important time lag for a region's independent variables, these specific time-lagged predictors were included as predictors in the candidate models, otherwise time-lags on predictors were not evaluated. For any region(s) that showed support for changepoints in their temporal structure, the periods separated by those breaks were included as an interaction term on all other predictors. In the absence of any identified changepoints, no multi-year periods were included as potential regime impacts. All candidate models for a region were evaluated using AIC and AIC weights to rank the most parsimonious models for each region's dynamics.```{r}#| label: regression-analysis-data# Use scaled drivers for the regressions# Need to drop zooplankton because they are non-unique since we split the MAB EPU and repeated itregression_df <- drivers_scaled %>%select(!ends_with(c("zp_small","zp_large")))# Build back out the metadata:# Identify what region is associated with both: spectrum values & driver valuesregression_df <- regression_df %>%pivot_longer(names_to ="spectra_param", values_to ="spectra_values", cols =ends_with("slope") |ends_with("int")) %>%pivot_longer(names_to ="driver_var", values_to ="driver_values", cols =-matches("spectra|year")) %>%mutate(year =as.numeric(year),# A. Flag what the driver type wasdriver_type =case_when(str_detect(driver_var, "landings") ~"Commercial Landings",str_detect(driver_var, "sst") ~"SST",str_detect(driver_var, "index") ~"GSI",TRUE~"Missed Something"),# B. Flag what the spectrum feature wasparam_feature =case_when(# Flag what the Size Distribution Parameter wasstr_detect(spectra_param, "int") ~"Spectra Intercept",str_detect(spectra_param, "isd") ~"ISD Exponent",str_detect(spectra_param, "slope") ~"Spectra Slope",TRUE~"Missed Something"),# C. These are the areas associated with the Spectra Featuresspectra_area =case_when(str_detect(spectra_param, "All") ~"All",str_detect(spectra_param, "Georges") ~"Georges Bank",str_detect(spectra_param, "Gulf") ~"Gulf of Maine",str_detect(spectra_param, "Southern") ~"Southern New England",str_detect(spectra_param, "Mid-Atlantic") ~"Mid-Atlantic Bight"),spectra_area =factor(spectra_area, levels =c("All", "Gulf of Maine", "Georges Bank", "Southern New England", "Mid-Atlantic Bight")),# D. These are the areas of the driversdriver_area =case_when(str_detect(driver_var, "All") ~"All",str_detect(driver_var, "Georges") ~"Georges Bank",str_detect(driver_var, "Gulf") ~"Gulf of Maine",str_detect(driver_var, "Southern") ~"Southern New England",str_detect(driver_var, "Mid-Atlantic") ~"Mid-Atlantic Bight"),driver_area =factor(driver_area, levels =c("All", "Gulf of Maine", "Georges Bank", "Southern New England", "Mid-Atlantic Bight")),# E. Add Decade indecade =floor_decade(year),##regime = ifelse(year < 2010, "regime_1", "regime_2" ) ) # Filter it so predictors are only within matching regions, # or just overarching (GSI)regression_df <- regression_df %>%filter(driver_area == spectra_area | driver_type =="GSI") %>%arrange(year, driver_var)# Last wrangling:# columns for: year, slope, isd, spectra region, drver_region, GSI, sst_anom, landings# GSI is so annoying here, just rejoin it without the area columnregression_df <- regression_df %>%select(-c(spectra_param, driver_var)) %>%pivot_wider(names_from ="driver_type", values_from ="driver_values") %>%pivot_wider(names_from ="param_feature", values_from ="spectra_values") %>%select(-GSI) %>%filter(driver_area !="All") %>%left_join( select(clim_idx, year, GSI = value), by =c("year") ) %>%mutate(spectra_area =fct_drop(spectra_area))# Add zooplankton here so it can be checked in the regressions# Zooplankton values are duplicated in MAB and SNE b/c its the same EPUregression_df <-left_join( regression_df, zp_index %>%rename(spectra_area = survey_area), by =join_by(year, spectra_area))```# Results### Abundance Changes```{r}#| label: abundance and biomass summary numbers#---- Overall Abundance Trends# Total abundance and biomass for the whole shelfshelf_summary <- catch_size_bins %>%group_by(Year) %>%summarise(lwbio_mill =sum(strat_total_lwbio_s, na.rm = T) /1e6,abund_mill =sum(strat_total_abund_s, na.rm = T) /1e6)# Total Abundance and Biomass for the regionsregion_summary <- catch_size_bins %>%left_join(area_df) %>%group_by(Year, area_titles) %>%summarise(lwbio_mill =sum(strat_total_lwbio_s, na.rm = T) /1e6,abund_mill =sum(strat_total_abund_s, na.rm = T) /1e6,.groups ="drop") %>%mutate(area_titles =factor(area_titles, levels = area_levels_long))# Total abundance and biomass for functional Groups within the regionsfunctional_group_summary <- catch_size_bins %>%left_join(area_df) %>%group_by(Year, area_titles, hare_group) %>%summarise(lwbio_mill =sum(strat_total_lwbio_s, na.rm = T) /1e6,abund_mill =sum(strat_total_abund_s, na.rm = T) /1e6,.groups ="drop") %>%mutate(area_titles =factor(area_titles, levels = area_levels_long))```Stratified abundance estimates have been gradually increasing on the Northeast Shelf since 1970. Abundance growth accelerated after 2007 to peak levels in 2014. Abundance then began declining, which continued through 2019 (Figure 2).```{r}#| label: abundance totals by area#| eval: true#| fig.width: 8#| fig-height: 6# Abundance 1.shelf_abund_p <- shelf_summary %>%mutate(area_titles ="Northeast Shelf") %>%ggplot() +geom_line(aes(Year, abund_mill), color =gmri_cols("dark gray"), linewidth =1) +#geom_col(aes(Year, abund_mill), fill = gmri_cols("dark gray")) +scale_y_continuous(labels =label_comma()) +facet_wrap(~area_titles) +labs(y ="Abundance (millions)")# Abundance 2.region_abund_p <- region_summary %>%ggplot() +geom_col(aes(Year, abund_mill, fill = area_titles), show.legend = F) +scale_fill_gmri() +scale_y_continuous(labels =label_comma()) +facet_wrap(~area_titles, ncol =1, scales ="free") +labs(y ="Abundance (millions)")shelf_abund_p | region_abund_p```In the Gulf of Maine, fish abundance remained relatively-low until the 1990's, at which point it began to steadily rise--closely aligning with the pattern for the Northeast Shelf. This increase in abundance reversed briefly during 2002-2006, but continued to rise after 2006 before hitting a regional peak in 2016. Populations then declined through 2019 similar to the shelf-wide trend. Georges Bank abundances were consistently low from 1970 around 2010. By 2014 abundance had roughly quadrupled, propelled by strong recruitment classes of haddock. After 2014, abundance quickly fell to numbers more similar to the 1970-2010 levels by the end of the decade. Community abundance in Southern New England displayed higher inter-annual changes across all years compared to both Gulf of Maine and Georges Bank. Abundance in this area showed a less dramatic rise and fall than the Northern regions. Abundances began increasing rapidly here in 2007, before falling back to earlier levels by the end of the 2010's decade. The Mid Atlantic Bight displayed the most inter-annual variability and had no major trends in fish abundance.#### Regional Abundances::: panel-tabset##### Northeast Shelf```{r}#| fig.width: 8#| fig-height: 6shelf_fgroup_summary <- catch_size_bins %>%mutate(area_titles ="Northeast Shelf") %>%group_by(Year, area_titles, hare_group) %>%summarise(lwbio_mill =sum(strat_total_lwbio_s, na.rm = T) /1e6,abund_mill =sum(strat_total_abund_s, na.rm = T) /1e6,.groups ="drop")# Make Plot for Functional Groupsshelf_fgroup_abund_p <- shelf_fgroup_summary %>%ggplot() +geom_col(aes(Year, abund_mill, fill = hare_group), show.legend = F) +scale_fill_gmri() +scale_y_continuous(labels =label_comma()) +facet_wrap(~hare_group, ncol =1, scales ="free") +labs(y ="Abundance (millions)")# Place Functional Groups Beside Shelfshelf_abund_p | shelf_fgroup_abund_p```##### Gulf of Maine```{r}#| fig.width: 8#| fig-height: 6area_option <-"Gulf of Maine"plot_area_abund_summary <-function(area_option){ area_abund_p <- region_summary %>%filter(area_titles == area_option) %>%ggplot() +geom_line(aes(Year, abund_mill), color =gmri_cols("dark gray"), linewidth =1) +#geom_col(aes(Year, abund_mill), fill = gmri_cols("dark gray")) +scale_y_continuous(labels =label_comma()) +facet_wrap(~area_titles) +labs(y ="Abundance (millions)")# Make Plot for Functional Groupsarea_fgroup_abund_p <- functional_group_summary %>%filter(area_titles == area_option) %>%ggplot() +geom_col(aes(Year, abund_mill, fill = hare_group), show.legend = F) +scale_fill_gmri() +scale_y_continuous(labels =label_comma()) +facet_wrap(~hare_group, ncol =1, scales ="free") +labs(y ="Abundance (millions)")# Place Functional Groups Beside Shelfp_out <- area_abund_p | area_fgroup_abund_preturn(p_out)}plot_area_abund_summary(area_option ="Gulf of Maine")```##### Georges Bank```{r}#| fig.width: 8#| fig-height: 6plot_area_abund_summary(area_option ="Georges Bank")```##### Southern New England```{r}#| fig.width: 8#| fig-height: 6plot_area_abund_summary(area_option ="Southern New England")```##### Mid-Atlantic Bight```{r}#| fig.width: 8#| fig-height: 6plot_area_abund_summary(area_option ="Mid-Atlantic Bight")```:::Groundfish species were the dominant functional group driving the abundance and biomass trends in the Gulf of Maine and Georges Bank, with the two southern regions showing a more balanced abundance distribution among the five functional groups (Figure 2).#### Total Abundance by Individual Body-Size```{r}#| label: bodymass biomass distributions#| fig.width: 10#| fig-height: 8# panel plot the biomass by body sizeplot_abundance_vertical <-function(p_area ="GoM", show_facets = F, ylab ="Abundance (million)"){ p <- fgroup_area$weight_bins %>%left_join(area_df) %>%filter(survey_area == p_area) %>%complete(Year, hare_group, area_titles, left_lim) %>%ggplot(aes(Year, strat_abund_mill, fill = left_lim)) +geom_col(color ="gray30", linewidth =0.1) +facet_grid(hare_group~area_titles, scales ="free") +scale_fill_distiller(palette ="RdYlGn", direction =-1, breaks =seq(0,14, 2), labels =math_format(2^.x)) +guides(fill =guide_colorbar(title.position ="top", title.hjust =0.5, barwidth =unit(4, "in"),frame.colour ="black", ticks.colour ="black")) +scale_y_continuous(labels =comma_format()) +labs(y = ylab, x =NULL, fill ="Minimum Individual Weight 2^x (g)") +theme(legend.position ="bottom",axis.text.x =element_text(angle =45, hjust =1),plot.margin =margin(t =0, r =0, b =0, l =2))if(show_facets == F){p <- p +theme(strip.text.y =element_blank(), strip.background.y =element_blank())}return(p)}# Build the piecesshelf_abunds <-plot_abundance_vertical("Northeast Shelf", show_facets = F)gom_abunds <-plot_abundance_vertical("GoM", F, ylab =NULL)gb_abunds <-plot_abundance_vertical("GB", F, ylab =NULL)sne_abunds <-plot_abundance_vertical("SNE", F, ylab =NULL)mab_abunds <-plot_abundance_vertical("MAB", T, ylab =NULL)# patch together( shelf_abunds | gom_abunds |gb_abunds |sne_abunds |gb_abunds |mab_abunds ) +plot_layout(guides ="collect")```### Biomass Changes```{r}#| label: biomass totals by area#| fig.width: 8#| fig-height: 6#----- Overall Biomass# Repeat for Biomass# Abundance 1.shelf_bio_p <- shelf_summary %>%mutate(area_titles ="Northeast Shelf") %>%ggplot() +geom_line(aes(Year, lwbio_mill), color =gmri_cols("dark gray"), linewidth =1) +#geom_col(aes(Year, abund_mill), fill = gmri_cols("dark gray")) +scale_y_continuous(labels =label_comma()) +facet_wrap(~area_titles) +labs(y ="Biomass (million kg)")# Abundance 2.region_bio_p <- region_summary %>%ggplot() +geom_col(aes(Year, lwbio_mill, fill = area_titles), show.legend = F) +scale_fill_gmri() +scale_y_continuous(labels =label_comma()) +facet_wrap(~area_titles, ncol =1, scales ="free") +labs(y ="Biomass (million kg)")shelf_bio_p | region_bio_p```Similar to abundance, the overall biomass was highest in the two northern regions, the Gulf of Maine and Georges Bank. Roughly half of the biomass sampled in these regions can be attributed to groundfish species, with the second largest contributions coming from elasmobranchs. Within the groundfish biomass, larger individuals \>2kg in particular, declined during the 70's and 80's in these regions, never truly recovering. Beginning in the 2000's there were signs that groundfish abundances were increasing as evidenced by increasing numbers of smaller individuals, however in both regions this trend appears to have reversed by the mid 2010's. Elasmobranch biomass increased steadily throughout the survey time period across all regions, with the exception of southern New England. This area showed large 5-10 year swings in biomass, but no clear long-term trend. Larger elasmobranch were rare in all regions except for a period spanning the late 70's through the early 90's isolated to Georges Bank. Demersal species biomass was highest in the Gulf of Maine, dwarfing their contributions in other regions. Their biomass declined in the 70's, was flat until the late 90's, remaining relatively high until declining in the late 2010's. Pelagic species biomass was low in all regions, and is unlikely to be representative of true biomass trends due to gear selectivity.#### Regional BiomassThere was a distinct difference between Northern and Southern regions in the way biomass was distributed among the different functional groups. The largest contributors to biomass in the southern regions (southern New England & mid-Atlantic bight) was the elasmobranch community. While the northern regions (Gulf of Maine & Georges Bank) each had similar quantities of elasmobranch biomasses, there was also a comparable contribution of groundfish and in the Gulf of Maine there was a major component of demersal species as well.::: panel-tabset##### Northeast Shelf```{r}#| fig.width: 8#| fig-height: 6# Make Plot for Functional Groupsshelf_fgroup_bio_p <- shelf_fgroup_summary %>%ggplot() +geom_col(aes(Year, lwbio_mill, fill = hare_group), show.legend = F) +scale_fill_gmri() +scale_y_continuous(labels =label_comma()) +facet_wrap(~hare_group, ncol =1, scales ="free") +labs(y ="Biomass (million kg)")# Place Functional Groups Beside Shelfshelf_bio_p | shelf_fgroup_bio_p```##### Gulf of Maine```{r}#| fig.width: 8#| fig-height: 6area_option <-"Gulf of Maine"plot_area_bio_summary <-function(area_option){ area_bio_p <- region_summary %>%filter(area_titles == area_option) %>%ggplot() +geom_line(aes(Year, lwbio_mill), color =gmri_cols("dark gray"), linewidth =1) +scale_y_continuous(labels =label_comma()) +facet_wrap(~area_titles) +labs(y ="Biomass (million kg)")# Make Plot for Functional Groupsarea_fgroup_bio_p <- functional_group_summary %>%filter(area_titles == area_option) %>%ggplot() +geom_col(aes(Year, lwbio_mill, fill = hare_group), show.legend = F) +scale_fill_gmri() +scale_y_continuous(labels =label_comma()) +facet_wrap(~hare_group, ncol =1, scales ="free") +labs(y ="Biomass (million kg)")# Place Functional Groups Beside Shelfp_out <- area_bio_p | area_fgroup_bio_preturn(p_out)}plot_area_bio_summary(area_option ="Gulf of Maine")```##### Georges Bank```{r}#| fig.width: 8#| fig-height: 6plot_area_bio_summary(area_option ="Georges Bank")```##### Southern New England```{r}#| fig.width: 8#| fig-height: 6plot_area_bio_summary(area_option ="Southern New England")```##### Mid-Atlantic Bight```{r}#| fig.width: 8#| fig-height: 6plot_area_bio_summary(area_option ="Mid-Atlantic Bight")```:::#### Total Biomass by Individual Body-Size```{r}#| label: bodymass abundance distributions#| fig.width: 10#| fig-height: 8# panel plot the biomass by body sizeplot_bodymass_vertical <-function(p_area ="GoM", show_facets = F, ylab ="Biomass (million kg)"){ p <- fgroup_area$weight_bins %>%left_join(area_df) %>%filter(survey_area == p_area) %>%complete(Year, hare_group, area_titles, left_lim) %>%ggplot(aes(Year, strat_lwbio_mill, fill = left_lim)) +geom_col(color ="gray30", linewidth =0.1) +facet_grid(hare_group~area_titles, scales ="free") +scale_fill_distiller(palette ="RdYlGn", direction =-1, breaks =seq(0,14, 2), labels =math_format(2^.x)) +guides(fill =guide_colorbar(title.position ="top", title.hjust =0.5, barwidth =unit(4, "in"),frame.colour ="black", ticks.colour ="black")) +scale_y_continuous(labels =comma_format()) +labs(y = ylab, x =NULL, fill ="Minimum Individual Weight 2^x (g)") +theme(legend.position ="bottom",axis.text.x =element_text(angle =45, hjust =1),plot.margin =margin(t =0, r =0, b =0, l =2))if(show_facets == F){p <- p +theme(strip.text.y =element_blank(), strip.background.y =element_blank())}return(p)}# Build the piecesshelf_bmass <-plot_bodymass_vertical("Northeast Shelf", show_facets = F)gom_bmass <-plot_bodymass_vertical("GoM", F, ylab =NULL)gb_bmass <-plot_bodymass_vertical("GB", F, ylab =NULL)sne_bmass <-plot_bodymass_vertical("SNE", F, ylab =NULL)mab_bmass <-plot_bodymass_vertical("MAB", T, ylab =NULL)# patch together( shelf_bmass | gom_bmass |gb_bmass |sne_bmass |gb_bmass |mab_bmass ) +plot_layout(guides ="collect")```### Body-Size TrendsThe average fish size in the Gulf of Maine (length and weight) declined the greatest of all regions over our study period. The average individual length was greatest in the 1970's in the 35-40cm range, falling to 28-33cm over the last decade. Body-weight fell dramatically in the 1980's, from around .75kg in the 1970's to .25-.30kg, roughly a third of what it had been. Georges Bank body sizes also declined during the study period, but less dramatically. Both of these Northern regions had brief period in the late 2000's where average length and weight rose, before falling again in the 2010's. The MAB region was the only region to see a long-term increase in both length and weight during the study period. SNE saw no long-term change in length, and a minor decline in average body-weight.::: panel-tabset#### All Species```{r}#| label: regional-size-trends#| fig.height: 8# Re-factorregional_size <- regional_size %>%filter(Year <2020) %>%mutate(survey_area =factor(survey_area, area_levels),area_copy = survey_area)# Get the average length and weight at each decadeavg_size_decades <- regional_size %>%mutate(decade_left =as.numeric(as.character(floor_decade(Year))),decade_right =as.numeric(as.character(floor_decade(Year+10))) -1,decade_center = (decade_left + decade_right)/2,decade_lab =str_c(decade_left, "-", decade_right)) %>%group_by(area_titles, decade_left, decade_right, decade_center, decade_lab) %>%summarise(mean_len_cm =mean(mean_len_cm),mean_wt_kg =mean(mean_wt_kg),.groups ="drop" )# Length plotavg_len_p <- regional_size %>%ggplot() +geom_line(aes(Year, mean_len_cm, group = area_titles), linewidth =1,show.legend = F, color ="gray60") +geom_segment(data = avg_size_decades,aes(x = decade_left, xend = decade_right, y = mean_len_cm, yend = mean_len_cm), linewidth =1.5) +geom_text(data = avg_size_decades,aes(x = decade_center, y = mean_len_cm, label =str_c(round(mean_len_cm, 1)) ), vjust =-2.25, size =4, fontface ="bold") +scale_color_gmri() +facet_wrap(~area_titles, ncol =1, scales ="free") +scale_y_continuous(expand =expansion(mult =c(0.1, 0.55))) +labs(title ="Average Length",y ="Length (cm)",color ="Region")# Weight plotavg_wt_p <- regional_size %>%ggplot(aes(Year, mean_wt_kg)) +geom_line(aes(group = area_titles), linewidth =1,show.legend = F,color ="gray60") +geom_segment(data = avg_size_decades,aes(x = decade_left, xend = decade_right, y = mean_wt_kg, yend = mean_wt_kg), linewidth =1.5) +geom_text(data = avg_size_decades,aes(x = decade_center, y = mean_wt_kg, label =str_c(round(mean_wt_kg, 1)) ), vjust =-2.25, size =4, fontface ="bold") +scale_color_gmri() +scale_y_continuous(expand =expansion(mult =c(0.1, 0.55))) +facet_wrap(~area_titles, ncol =1, scales ="free") +labs(title ="Average Weight",y ="Weight (kg)",color ="Region")# Plot side by side(avg_len_p | avg_wt_p) ```#### Functional Group Avg. Length```{r}#| label: functional-group-length-trends#| fig-height: 6# Get the average length and weight at each decadeavg_fgroup_decades <- functional_group_size %>%mutate(decade_left =as.numeric(as.character(floor_decade(Year))),decade_right =as.numeric(as.character(floor_decade(Year+10))) -1,decade_center = (decade_left + decade_right)/2,decade_lab =str_c(decade_left, "-", decade_right)) %>%group_by(area_titles, hare_group, decade_left, decade_right, decade_center, decade_lab) %>%summarise(mean_len_cm =mean(mean_len_cm),mean_wt_kg =mean(mean_wt_kg),.groups ="drop")# Plot the Length changesfunctional_group_size %>%ggplot() +geom_line(aes(Year, mean_len_cm, group = area_titles), linewidth =1,show.legend = F, color ="gray60") +geom_segment(data = avg_fgroup_decades,aes(x = decade_left, xend = decade_right, y = mean_len_cm, yend = mean_len_cm), linewidth =1.5) +geom_text(data = avg_fgroup_decades,aes(x = decade_center, y = mean_len_cm, #y = Inf,label =str_c(round(mean_len_cm, 1)) ), vjust =-2.5, #vjust = 1.75,size =3.5, fontface ="bold") +scale_y_continuous(expand =expansion(mult =c(0.1, 0.4))) +scale_color_gmri() +facet_grid(hare_group~area_titles, scales ="free_y") +labs(title ="Average Length",y ="Length (cm)",color ="Region")```#### Functional Group Avg. Weight```{r}#| label: functional-group-weight-trends#| fig-height: 6# Average Weightsfunctional_group_size %>%ggplot() +geom_line(aes(Year, mean_wt_kg, group = area_titles), linewidth =1,show.legend = F, color ="gray60") +geom_segment(data = avg_fgroup_decades,aes(x = decade_left, xend = decade_right, y = mean_wt_kg, yend = mean_wt_kg), linewidth =1.5) +geom_text(data = avg_fgroup_decades,aes(x = decade_center, y = mean_wt_kg, label =str_c(round(mean_len_cm, 1)) ), vjust =-2.5, size =3.5, fontface ="bold") +scale_y_continuous(expand =expansion(mult =c(0.1, 0.4))) +scale_color_gmri() +facet_grid(hare_group~area_titles, scales ="free_y") +labs(title ="Average Weight",y ="Weight (kg)",color ="Region")```:::### Regional Size SpectraAt the start of our time series, back in the 1970's, there was a clear difference in the relative positions of spectra parameters among the different regions. Gulf of Maine and Georges Bank showed the least steep spectra slopes in the earlier time periods with slopes around -1 & -1.1 respectively. The relatively flat slopes in these regions both steepened over time, settling near -1.3 (GoM) and -1.5 (GB). Gulf of Maine experienced much of its decline during the 1980's and 1990's. There was a brief reversal in this trend during the 2000's, but slopes continued to steepen by 2010 and remained steep through 2019. Georges Bank did not experience as rapid of a decline, but experienced a similar long-term steepening. In contrast to the northern regions, SNE and MAB had steeper slopes in the -1.2 to -1.5 territory. The long term pattern for SNE was one of increasing volatility, but not so much a decline. The spectra slope for the MAB was less volatile, but similarly maintained a relatively stable wander around -1.4. By the end of the study period all regions had slopes that were at or near a similar level.```{r}#| label: size spectra results# Make decadal averages of spectra slopesindices_decades <- region_indices %>%mutate(Year =as.numeric(Year),decade_left =as.numeric(as.character(floor_decade(Year))),decade_right =as.numeric(as.character(floor_decade(Year +10))) -1,decade_center = (decade_left + decade_right)/2,decade_lab =str_c(decade_left, "-", decade_right)) %>%group_by(area_titles, decade_left, decade_right, decade_center, decade_lab) %>%summarise( mean_b =mean(b, na.rm = T),mean_l2slope =mean(log2_slope_strat, na.rm = T), .groups ="drop")```::: panel-tabset#### Exponent of Size Spectra```{r}#| fig.height: 7# 1. Exponent of ISD, MLE methodisd_timeline <- region_indices %>%ggplot(aes(yr, b, group = area_titles)) +geom_line(aes(group = area_titles), color ="gray60", linewidth =1) +geom_segment(data = indices_decades,aes(x = decade_left, xend = decade_right, y = mean_b, yend = mean_b), linewidth =1.5) +geom_text(data = indices_decades,aes(x = decade_center, y = mean_b, label =str_c(round(mean_b, 2)) ), vjust =-2.25, size =4, fontface ="bold") +scale_y_continuous(expand =expansion(mult =c(0.1, 0.65))) +# scale_y_continuous(expand = expansion(add = c(0,.1))) +scale_color_gmri() +facet_wrap(~area_titles, ncol =1, scales ="free") +labs(x ="Year", y ="Abundance Size Spectrum Slope (b)")isd_timeline # #-------------# # # Build them separately# plot_spectra <- function(area_lab){# line_dat <- region_indices %>% # filter(area_titles == area_lab)# decade_dat <- indices_decades %>% # filter(area_titles == area_lab)# # line_dat %>% # ggplot(aes(yr, b, group = area_titles)) +# geom_line(# aes(group = area_titles), # color = "gray60", # linewidth = 1) +# geom_segment(# data = decade_dat,# aes(x = decade_left, # xend = decade_right, # y = mean_b, # yend = mean_b), # linewidth = 1.5) +# geom_text(# data = decade_dat,# aes(x = decade_center, # y = mean_b, # label = str_c(round(mean_b, 2)) ), # vjust = -2.25, # size = 4, fontface = "bold") +# scale_y_continuous(expand = expansion(mult = c(0.1, 0.55))) +# scale_color_gmri() +# facet_wrap(~area_titles, ncol = 1, scales = "free") +# labs(x = NULL, y = NULL, caption = NULL, title = NULL, subtitle = NULL) +# theme(plot.margin = margin(t = 0, b = 0))# }# # # Just the GOM plot needs some added features# # # Just manually do gom...# line_dat_gom <- region_indices %>% # filter(area_titles == "Gulf of Maine") %>% # mutate(regime_col = ifelse(# yr < 1999, gmri_cols("green"),# ifelse(yr > 2006, gmri_cols("orange"),# gmri_cols("blue"))))# decade_dat_gom <- indices_decades %>% # filter(area_titles == "Gulf of Maine")# # gom_spectra <- line_dat_gom %>% # ggplot(aes(yr, b, group = area_titles)) +# geom_line(# aes(group = area_titles,# color = I(regime_col)), # linewidth = 1, show.legend = F) +# geom_segment(# data = decade_dat_gom,# aes(x = decade_left, # xend = decade_right, # y = mean_b, # yend = mean_b), # linewidth = 1.5) +# geom_text(# data = decade_dat_gom,# aes(x = decade_center, # y = mean_b, # label = str_c(round(mean_b, 2)) ), # vjust = -2.25, # size = 4, fontface = "bold") +# scale_y_continuous(expand = expansion(mult = c(0.1, 0.55))) +# scale_color_gmri() +# facet_wrap(~area_titles, ncol = 1, scales = "free") +# labs(x = NULL, y = NULL, caption = NULL, title = NULL, subtitle = NULL) +# theme(plot.margin = margin(t = 0, b = 0))# # # Then do the others# shelf_spectra <- plot_spectra(area_lab = "Northeast Shelf")# gb_spectra <- plot_spectra(area_lab = "Georges Bank")# mab_spectra <- plot_spectra(area_lab = "Mid-Atlantic Bight")# sne_spectra <- plot_spectra(area_lab = "Southern New England")# # shelf_spectra /# gom_spectra /# (gb_spectra +# labs(y = "Size Spectra Slope (b)")) /# sne_spectra /# mab_spectra +# labs(x = "Year")```#### Log2 Binned Size Spectra```{r}#| fig.height: 7# 2. Plot the Regional Slopes, using the binned methodslope_timeline <- region_indices %>%ggplot(aes(yr, log2_slope_strat, group = area_titles)) +geom_line(aes(group = area_titles), linewidth =1, color ="gray70") +geom_segment(data = indices_decades,aes(x = decade_left, xend = decade_right, y = mean_l2slope, yend = mean_l2slope), linewidth =1.5) +geom_text(data = indices_decades,aes(x = decade_center, y = mean_l2slope, label =str_c(round(mean_l2slope, 1)) ), vjust =-1.05, size =3.5, fontface ="bold") +scale_y_continuous(expand =expansion(add =c(0,.1))) +scale_color_gmri() +facet_wrap(~area_titles, ncol =1) +labs(x ="Year", y ="Normalized Abundance Spectrum Slope") slope_timeline #+ geom_smooth(color = "orange", se = FALSE)```#### Size-Bin Changes```{r}# Take the data for each size bin# Standardize it against the all-year average# plot how it changed through time using geom_tile# Take all the data:# Get the average to normalize againstspectra_heatmap_data <- catch_size_bins %>%mutate(area_titles ="Northeast Shelf") %>%bind_rows(catch_size_bins) %>%split(.$area_titles) %>%map_dfr(function(x){ x %>%group_by(area_titles, Year, left_lim, bin_label, bin_width) %>% dplyr::summarize(single_year_total_abund =sum(strat_total_abund_s, na.rm = T),.groups ="drop") %>%group_by(area_titles, left_lim, bin_label, bin_width) %>%mutate(all_year_total =sum(single_year_total_abund),all_year_avg =mean(single_year_total_abund, na.rm = T),all_year_sd =sd(single_year_total_abund, na.rm = T)) %>%ungroup() %>%mutate(single_year_z = (single_year_total_abund - all_year_avg) / all_year_sd) }, .id ="area_titles") %>%mutate(area_titles =factor(area_titles, levels = area_levels_long))# Plot the heatmap:spectra_heatmaps <-ggplot(spectra_heatmap_data, aes(x = Year, y =2^left_lim, fill = single_year_z)) +geom_tile() +geom_hline(aes(yintercept = ((2^left_lim) + (2^(left_lim+1)))/2), color ="white", linewidth =0.3) +scale_x_continuous(expand =expansion(add =c(0.5,0.5))) +scale_y_continuous(trans ="log2",labels =label_log(base =2),breaks =2^seq(0,12, by =2),expand =expansion(add =c(0.5, 0.5))) +scale_fill_distiller(limits =c(-2,2), labels =c("<-2", c(-1:1), ">2"),oob = oob_squish, palette ="RdBu") +facet_wrap(~area_titles, ncol =1) +theme(legend.title =element_text(angle=90, size =12),legend.position ="right", legend.direction ="vertical") +guides(fill =guide_colorbar(title.position ="right",title.hjust =0.5,frame.colour ="black", ticks.colour ="black",barwidth =unit(1, "cm"),barheight =unit(8, "cm"))) +labs(y ="Body Weight (g)",fill ="Size-Specific Relative Abundance (z)")isd_timeline | spectra_heatmaps############################ # Question:# # Should I normalize the abundances by bin-width first# # Then show how abundances shifted on that all-sizes scale?# maybe# # Yea probably, jk... magnitudes are way off# Maybe within a year?# This code and its output should be# norm_test <- catch_size_bins %>% # mutate(area_titles = "Northeast Shelf") %>% # bind_rows(catch_size_bins) %>% # split(.$area_titles) %>% # map_dfr(function(x){# x %>% # group_by(area_titles, Year, left_lim, bin_label, bin_width) %>%# dplyr::summarize(# single_year_total_abund = sum(strat_total_abund_s, na.rm = T),# .groups = "drop") %>%# # Do we normalize first and then ignore the bins?# mutate(# single_year_norm = single_year_total_abund / bin_width# ) %>%# # Now I'm confused what to average and standardize against...# # I think I want the normalized abundance# # to be scaled against the average normalized abundance and variance across all years# group_by(area_titles) %>%# mutate(# all_size_avg = mean(single_year_norm, na.rm = T),# all_size_sd = sd(single_year_norm, na.rm = T)) %>%# ungroup() %>%# mutate(single_year_z = (single_year_norm - all_size_avg) / all_size_sd)# }, .id = "area_titles") %>% # mutate(area_titles = factor(area_titles, levels = area_levels_long))# # # # Norm test# ggplot(norm_test,# aes(x = Year, y = 2^left_lim,# fill = single_year_z)) +# geom_tile() +# geom_hline(aes(yintercept = ((2^left_lim) + (2^(left_lim+1)))/2), color = "white", linewidth = 0.3) +# scale_x_continuous(expand = expansion(add = c(0.5,0.5))) +# scale_y_continuous(# trans = "log2",# labels = label_log(base = 2),# breaks = 2^seq(0,12, by = 2),# expand = expansion(add = c(0.5, 0.5))) +# scale_fill_distiller(# limits = c(-2,2), # labels = c("<-2", c(-1:1), ">2"),# oob = oob_squish, palette = "RdBu") +# facet_wrap(~area_titles, ncol = 1) +# theme(# legend.title = element_text(angle= 90, size = 12),# legend.position = "right", # legend.direction = "vertical") +# guides(# fill = guide_colorbar(# title.position = "right",# title.hjust = 0.5,# frame.colour = "black", ticks.colour = "black",# barwidth = unit(1, "cm"),# barheight = unit(8, "cm"))) +# labs(y = "Body Weight (g)",# fill = "Change in All-Size Normalized Abundance (z)")```#### Surface Checking```{r}#| eval: false#| label: surfaces check# Plot the surface using the actual abundances#### Surface Plotsspectra_surface <-function(log2_totals, plot_title){# Pull data for matrix dat <- dplyr::select(log2_totals, left_lim, Year, norm_strat_abund)# Make matrix to plot spectrum_mat <-as.matrix(xtabs(norm_strat_abund~left_lim + Year, data = dat))# Make a flat surface of -1 slope?# Build plotly plot surf_plot <-plot_ly(x =as.numeric(colnames(spectrum_mat)), y =as.numeric(rownames(spectrum_mat)), z = spectrum_mat ) %>%add_surface() %>% plotly::layout(title =str_c(plot_title, " - Normalized Size Spectrum"),scene =list(xaxis =list(title ="Year", autorange ="reversed"),yaxis =list(title ="Log2( BodyMass (g) )"),zaxis =list(title ="Log2( Normalized Abundance) "),camera =list(eye =list(x =1, y =1, z =1.25)) ))#return plotreturn(surf_plot)}# Plot it using the normalized abundances# Get annual totals within each areaareas_aggregated <- catch_size_bins %>%split(.$area_titles) %>%map(function(area_data){ area_data %>%split(.$Year) %>%map_dfr(function(yr_data){aggregate_log2_bins(yr_data) %>%mutate(log2_bins =as.numeric(log2_bins),norm_strat_abund =log2(norm_strat_abund),norm_strat_abund =ifelse(is.na(norm_strat_abund), 0, norm_strat_abund)) }, .id ="Year") })# Interactive Plot?library(plotly)# Gulf of Mainespectra_surface(log2_totals =filter(areas_aggregated$`Gulf of Maine`, Year >=1970), plot_title ="Gulf of Maine")# Everywhereshelf_bin_agg <- catch_size_bins %>%split(.$Year) %>%map_dfr(function(yr_data){aggregate_log2_bins(yr_data) %>%mutate(log2_bins =as.numeric(log2_bins),norm_strat_abund =log2(norm_strat_abund),norm_strat_abund =ifelse(is.na(norm_strat_abund), 0, norm_strat_abund)) }, .id ="Year")spectra_surface(log2_totals =filter(shelf_bin_agg, Year >=2000),plot_title ="Northeast Shelf")#```:::## Size Spectra Drivers```{r}#| fig-height: 8d1 <- temp_regimes %>%select(survey_area, year = yr, idx = sst_anom) %>%mutate(idx_name ="SST\nAnomaly")d2 <- landings_summ %>%select(survey_area, year, idx = total_wt_z) %>%mutate(idx_name ="Commercial\nLandings")d3 <- zp_index %>%rename(`Small\nZooplankton`= zp_small,`Large\nZooplankton`= zp_large) %>%pivot_longer(cols =-c(year, survey_area), names_to ="idx_name", values_to ="idx")# Area-wide scaled4 <- clim_drivers %>%mutate(year = lubridate::year(Time)) %>%group_by(year) %>%summarise(val =mean(Value)) %>%select(year, idx = val) %>%mutate(idx_name ="Gulf Stream Index",survey_area ="Northeast Shelf")d_all <-bind_rows(list(d1,d2,d3,d4)) %>%mutate(survey_area =case_when( survey_area =="GoM"~"Gulf of Maine", survey_area =="GB"~"Georges Bank", survey_area =="MAB"~"Mid-Atlantic Bight", survey_area =="SNE"~"Southern New England",TRUE~ survey_area),survey_area =factor(survey_area, levels = area_levels_long),idx_name =factor(idx_name, levels =c("Gulf Stream Index","Commercial\nLandings","SST\nAnomaly","Small\nZooplankton","Large\nZooplankton")),anom_direction =ifelse(idx >0, "Positive", "Negative"))# Plot the over-arching onesshelf_drivers_p <- d_all %>%filter(survey_area =="Northeast Shelf", idx_name %in%c("Gulf Stream Index", "SST\nAnomaly")) %>%ggplot(aes(year, idx)) +geom_hline(yintercept =0, linewidth =0.5, lty =4) +geom_path(aes(color = anom_direction, group =1),linewidth =1.25, alpha =0.8, show.legend = F) +scale_x_continuous(limits =c(1970,2019),expand =expansion(add =c(0.25,0.25))) +scale_fill_gmri() +scale_color_gmri() +facet_grid(idx_name~survey_area) +labs(x ="Year", y ="Index Value")# Plot the regional onesregion_drivers_p <- d_all %>%filter(survey_area !="Northeast Shelf", idx_name !="Gulf Stream Index") %>%ggplot(aes(year, idx)) +geom_hline(yintercept =0, linewidth =0.5, lty =4) +geom_path(aes(color = anom_direction, group =1),linewidth =1.25, alpha =0.8, show.legend = F) +scale_x_continuous(limits =c(1970,2019),expand =expansion(add =c(0.25,0.25))) +scale_fill_gmri() +scale_color_gmri() +facet_grid(idx_name~ survey_area, scales ="free") +labs(x ="Year", y ="Index Value")# Annnnd assemble(shelf_drivers_p +theme(plot.margin =margin(t =0, b =0))) / ( region_drivers_p +theme(plot.margin =margin(t =0, b =0))) + patchwork::plot_layout(heights =c(2,3))```### Lagged Impacts of PredictorsExploratory analysis on potentially important predictor time lags through the use of cross correlation functions raised 6 potential lagged-driver candidates for the driver regression analyses. CCF identified potential time lags in Georges Bank predictors, and in the Mid-Atlantic Bight. Lagged predictors added to Georges Bank regression model selection process included: A 2-year lag on the small zooplankton index, a 1-year lag on the large zooplankton index, a 1-year lag of SST, 4-year lag & 1-year lags of the GSI, and 4 & 5 year lags of commercial landings. Lagged predictors added to Mid-Atlantic Bight regression model selection process included: 5 & 7 year lags on the small zooplankton index and a 2-year lag on SST.```{r}#| label: ccf-results#| fig-height: 8#| fig-width: 8# Run CCF function to get lagged correlations,# Put that data in one table to plot in a better comparison form#------------ Processing Data for CCF. ------------------------# Scale Everything Before CCF,# Reshape the Drivers Wide to scale# Scale the columns independently using the variance in all years, but within an area# Are the columns scaling independently or together?# Confirmed independently scaling# Drivers scaled trimmed is a matrix of all the different timeseries# Columns that end with slope or intercept are pulled out to get a column of the response variables (slope features)# Columns that don't contain "spectra" or "year" capture the remaining features# Next we pull out the region information for eachdriver_ccf_prep <- drivers_scaled %>%select(-All_sst_anom) %>%#rownames_to_column(var = "year") %>% pivot_longer(names_to ="spectra_param", values_to ="spectra_values", cols =ends_with("slope") |ends_with("int")) %>%pivot_longer(names_to ="driver_var", values_to ="driver_values", cols =-matches("spectra|year")) %>%mutate(# C. These are the areas associated with the Spectra Featuresspectra_area =case_when(str_detect(spectra_param, "All") ~"All",str_detect(spectra_param, "Georges") ~"Georges Bank",str_detect(spectra_param, "Gulf") ~"Gulf of Maine",str_detect(spectra_param, "Southern") ~"Southern New England",str_detect(spectra_param, "Mid-Atlantic") ~"Mid-Atlantic Bight"),spectra_area =factor(spectra_area, levels =c("All", "Gulf of Maine", "Georges Bank", "Southern New England", "Mid-Atlantic Bight")),# D. These are the areas of the driversdriver_area =case_when(str_detect(driver_var, "All") ~"All",str_detect(driver_var, "Georges") ~"Georges Bank",str_detect(driver_var, "Gulf") ~"Gulf of Maine",str_detect(driver_var, "Southern") ~"Southern New England",str_detect(driver_var, "Mid-Atlantic") ~"Mid-Atlantic Bight"),driver_area =factor(driver_area, levels =c("All", "Gulf of Maine", "Georges Bank", "Southern New England", "Mid-Atlantic Bight"))) # Filter it out so there is only cases where the driver area matchesdriver_ccf_prep <- driver_ccf_prep %>%filter(driver_area == spectra_area | driver_area =="All") %>%arrange(year, driver_var)# Walk through each xvariable, and see how it correlates with each yvar# split on xvar# then split on yvars# make sure order is goodccf_relationships <- driver_ccf_prep %>%drop_na() %>%split(.$spectra_param) %>%map_dfr(function(x_param){ x_param %>%split(.$driver_var) %>%map_dfr(function(driver_y_data){ ccf_df <-get_ccf_vector(x = driver_y_data$spectra_values,y = driver_y_data$driver_values ) }, .id ="driver_var") }, .id ="spectra_param")# Build back out the labels for plottingccf_plot_data <- ccf_relationships %>%mutate(# Flag what the driver type wasdriver_type =case_when(str_detect(driver_var, "landings") ~"Commercial Landings",str_detect(driver_var, "sst") ~"SST",str_detect(driver_var, "index") ~"GSI",str_detect(driver_var, "small") ~"ZP-S",str_detect(driver_var, "large") ~"ZP-L",TRUE~"Missed Something"),param_feature =case_when(# Flag what the Size Distribution Parameter wasstr_detect(spectra_param, "int") ~"Spectra Intercept",str_detect(spectra_param, "isd") ~"ISD Exponent",str_detect(spectra_param, "slope") ~"Spectra Slope",TRUE~"Missed Something"),spectra_region =case_when(# Flag what region the driver was coming fromstr_detect(spectra_param, "All") ~"All",str_detect(spectra_param, "Georges") ~"Georges Bank",str_detect(spectra_param, "Gulf") ~"Gulf of Maine",str_detect(spectra_param, "Southern") ~"Southern New England",str_detect(spectra_param, "Mid-Atlantic") ~"Mid-Atlantic Bight"),# Flag when it crosses thresholdsig_flag =ifelse(acf < signeg | acf > sigpos, T, F),# Set Factor Levelsspectra_region =factor(spectra_region, levels =c("All", "Gulf of Maine", "Georges Bank", "Southern New England", "Mid-Atlantic Bight")))# Limit to one Response:ccf_plot_data <-filter(ccf_plot_data, param_feature =="ISD Exponent") %>%filter(spectra_region !="All")# Plot themccf_plot_data %>%ggplot(aes(x = lag, y = acf, group = driver_var, color = driver_type, fill = driver_type)) +geom_col(alpha =0.65) +geom_col(data =filter(ccf_plot_data, sig_flag),aes(lag, acf, fill = driver_type, group = driver_var), color ="black") +geom_text(data =filter(ccf_plot_data, sig_flag),aes(lag, y =0, label = lag, vjust =ifelse(acf <0, 1.5,-1.5)), color ="black") +geom_line(aes(x = lag, y = sigpos), linetype =2, color ="gray25") +geom_line(aes(x = lag, y = signeg), linetype =2, color ="gray25") +geom_hline(yintercept =0, color ="gray25", linewidth =1) +geom_vline(xintercept =0, color ="black", linewidth =1) +scale_color_gmri() +scale_fill_gmri() +facet_grid(spectra_region~driver_type*param_feature) +scale_x_continuous(limits =c(-15, 15), breaks =seq(from =-15, to =12, by =1)) +theme(legend.position ="bottom",axis.text =element_text(size =8)) +labs(x ="Driver Lag", y ="Correlation (CCF)", fill ="Driver Type:", color ="Driver Type:",caption ="All predictors scaled over year range of 1982-2019. Lags exceeding significance threshold outlined and numbered.")```### Spectra Slope ChangepointsBreakpoint analyses showed support for ecological regime change in one region, the Gulf of Maine. Evidence of breakpoints were found in 1999 & 2007 -- when slope values increased briefly, before reversing course and falling further. The best-supported breakpoint model structure included breaks in the linear trend at 1999 & 2007, and a 2-year autocorrelation term. Suggesting that the community spectra behaved differently over 3 regimes with strong autocorrelation to its state two years prior. For Georges Bank spectra slopes showed a single trend structure, with slopes steepening over time. This was also the case for the Mid-Atlantic Bight, but with slopes becoming more shallow with time. In Southern New England a single mean (intercept) model best reflected the lack of change. With no strong support for either a long-term trend or break points. Based on these exploratory results, three regime periods were added to the Gulf of Maine's dataset for multiple regression analysis. The three regimes were 1982-1998, 1999-2006, 2007-2019. These regimes were included as an interaction term with each of the original hypothesized drivers, permitting potentially different influences of the drivers within each regime. The two-year lag on the spectrum slope was also included based on its support in the most-parsimonious changepoint analysis model. No breakpoints or autocorrelation terms were added to models for the other regions.```{r}#| label: regional changepoint analysis# Pull Regions to test independently:regression_df <- regression_df %>%rename(landings =`Commercial Landings`,b =`ISD Exponent`)gom_df <-filter(regression_df, spectra_area =="Gulf of Maine") gb_df <-filter(regression_df, spectra_area =="Georges Bank") %>%drop_na()sne_df <-filter(regression_df, spectra_area =="Southern New England") mab_df <-filter(regression_df, spectra_area =="Mid-Atlantic Bight") %>%drop_na()# Check each for different changepoint structuresgom_cpt <-envcpt(gom_df$b, verbose = F)gb_cpt <-envcpt(gb_df$b, verbose = F)sne_cpt <-envcpt(sne_df$b, verbose = F)mab_cpt <-envcpt(mab_df$b, verbose = F)# Pulling out details of best supported modelcpt_res <-list("gom"=list("cpt_res"= gom_cpt),"gb"=list("cpt_res"= gb_cpt),"sne"=list("cpt_res"= sne_cpt),"mab"=list("cpt_res"= mab_cpt)) %>%map(function(x){ best <-names(which.min(AIC(x$cpt_res))) best_summ <- x$cpt_res[[best]] x$best <- best x$best_summ <- best_summreturn(x)})# # Plot GOM, the only one with changepoints# plot(gom_cpt, main = "Gulf of Maine, Break Point Analysis")# Put the rest of the results into a delta AIC table:bind_rows(map(cpt_res, ~tibble("Most Supported Model"= .x[["best"]])), .id ="Region") %>%kable()```Change point analyses suggest there were as many as 3 distinct regimes in the Gulf of Maine slope changes. For all other regions there was not strong support for change points.## Spectra Slope Regression Analysis# WORKING HERE - Effect Reportingfix the coefficients so the interactions now reflect the net result::: panel-tabset### Gulf of MaineModel rankings using AICc & delta-AICc (Figure 5.) for the Gulf of Maine's size spectrum slope models best support a regression model containing two predictors with a stationary effect across all years, and two predictors whose effects across were allowed to vary across the regime periods identified by the changepoint analysis (non-stationary effects). The two stationary drivers impacts were the negative impact of the small zooplankton index (p = 0.011) and a negative impact from the two-year autocorrelation term highlighted by the exploratory changepoint analysis (p = 0.033). The inclusion of an interaction term between the three regimes (1982-1998, 1999-2006, & 2007-2019) and both the commercial landings index and the large zooplankton index allowed these drivers to differently impact size spectrum slope within the three regimes. Over those three periods the impact of landings during the first regime was positive (p = 0.009), the 1999-2006 regime was negative (p = 0.032), and during the third regime there was no relationship (p = 0.088). For the large zooplankton index during the first two regimes there was no significant impact (p = 0.2, p = \>0.9), however during the third regime large zooplankton had a positive effect on the spectrum slope (p = \<0.001). None of the terms for the regime periods themselves were significant in the final model. Models containing only stationary effects on the predictors were not retained by model selection. This agreed with visual assessments of their predictive performance and residual trends.```{r}#| label: gom-standard-models# change na. action for dredgeoptions(na.action ="na.fail")# Incorporate the patterns from the changepoint analysis# What if we enforce the break_periodsgom_df <-mutate(gom_df, regime =case_when( year <1999~"yrs_1982-1998", year <2007~"yrs_1999_2006", year >=2007~"yrs_2007_2019"))# Add the autoregressive predictors neatly/manuallygom_df_lag <- gom_df %>%mutate(blag1 = dplyr::lag(b, 1),blag2 = dplyr::lag(b, 2),zpslag1 = dplyr::lag(zp_small, 1),zpslag2 = dplyr::lag(zp_small, 2)) %>%drop_na(blag2)# kitchen sink Autoregressive modelgom_ar_models =lm( b ~ regime * SST + regime * GSI + regime * landings + regime * zp_small + regime * zp_large + regime * blag1 + regime * blag2, data = gom_df_lag)# Model selection on models with lags at 1+2 yearsresults_gom <-dredge(gom_ar_models)# Best autoregressive modelbest_gom <-get.models(results_gom, subset = delta ==0)[[1]]``````{r}#| label: gom-top-models-table# Okay quick issues, for non-stationary effects, maybe just take the models under delta = 2# and from them pull the summary# Pull details for models with delta AIC < 2top_mods_gom <-get.models(results_gom, subset = delta <2)# Peformance, Gets the AIC/AICc order top_perf <-map_dfr(top_mods_gom, function(x){glance(x) %>%mutate(AICc =AICc(x))}, .id ="id") %>%arrange(AICc)# Calculate delta AICctop_perf$Delta <- top_perf$AICc -min(top_perf$AICc)# Give the ranking based on that ordertop_perf <- top_perf %>%arrange(Delta) %>%mutate(`Model ID`=row_number(),`Model ID`=str_c("Model ", `Model ID`))# Retrieve Model Coefficients from "top" models# join in the model number based on AIC ordertop_coef <-map_dfr(top_mods_gom, tidy, .id ="id") %>%left_join(select(top_perf, id, `Model ID`)) %>%select(-id)# Reshape Performance Specs to bind_row latertop_perf <-select(top_perf, `Model ID`, r2 = r.squared, Delta) %>%pivot_longer(cols =-1, names_to ="term", values_to ="estimate") %>%mutate(fill_color ="lightgray")# Clean up the data and put it into a graphgom_table_data <- top_coef %>%# Tidy up the names and set the color scheme for fillmutate(term =ifelse(term =="landings", "Commercial Landings", term),term =ifelse(term =="zp_small", "Small Zooplankton", term),term =ifelse(term =="zp_large", "Large Zooplankton", term),term =ifelse(term =="blag2", "2-yr-AR", term),term =str_remove_all(term, "regime"),term =str_remove_all(term, "yrs_"),term =str_replace_all(term, ":", " ~ "),fill_color =ifelse(estimate<0, "#EACA00", "#00736D")) %>%# Join in the Performance Details and set Factor Ordersbind_rows(top_perf) %>%mutate(survey_area ="Gulf of Maine",`Model ID`=fct_rev(`Model ID`),estimate =round(estimate, 2)) %>%# Set the different types of estimates heremutate(type =case_when( term =="1999_2006"~"Regime\nIntercept", term =="2007_2019"~"Regime\nIntercept", term =="2-yr-AR"~"Fixed\nEffects", term =="r2"~"Model\nPerformance", term =="Delta"~"Model\nPerformance",str_detect(term," ~ ") ~"Regime\nInteraction",TRUE~"Fixed\nEffects"),type =factor(type, levels =c("Fixed\nEffects", "Regime\nIntercept", "Regime\nInteraction", "Model\nPerformance"))) #### Working HERE ###### At this point it is confusing how to read how# the regimes impact the fixed effects# We need to change that# How should we organize it?# Option 1: adjust the coefficients using the regime 1 coefficient + regime_x coefficients# then display it how we are here# Option 2: For the display, break Gulf of Maine into the three periods# Show how the predictions work####### Pull the regime parts out# Any terms that are involved in interactionsregime_coefs <- top_coef #%>% # split(.$`Model ID`) %>% # # everything here down will need to be in map()# pluck(1) %>% # filter(str_detect(term, ":|regime|landings|large"))# Pull the predictors for the primary period (factor level 1)p1_predictors <- regime_coefs %>%filter(!str_detect(term, ":")) %>%mutate(regime_period ="1982-1998") %>%select(-c(std.error, statistic, p.value))# Isolate the intercept adjustment from the regimesregime_adj <- p1_predictors %>%filter(str_detect(term, "regime")) %>%mutate(regime_period =str_remove_all(term, "regimeyrs_"),regime_period =str_replace(regime_period, "_", "-")) %>%select(-term) %>%pivot_wider(names_from ="regime_period", values_from ="estimate") %>%mutate(term ="(Intercept)")# Reshape the primary termsprimary_terms <- p1_predictors %>%pivot_wider(names_from ="regime_period", values_from ="estimate") %>%filter(!str_detect(term, "regime"))# Pull out the interactions and label the fixed effect and the regimeregime_interax <- regime_coefs %>%filter(str_detect(term, ":")) %>%mutate(regime_period =ifelse(str_detect(term, "1999"), "1999-2006", "2007-2019")) %>%separate(term, sep =":", remove = F, into =c("term_1", "term_2")) %>%mutate(term =ifelse(str_detect(term_1, "regime"), "zp_large", "landings")) %>%select(-c(std.error, statistic, p.value, term_1, term_2)) %>%pivot_wider(names_from ="regime_period", values_from ="estimate")# How the hell do I clean this part up?# Join the three pieces togetherprimary_terms %>%left_join(bind_rows(regime_adj, regime_interax)) %>%mutate(regime_2_net_change =`1982-1998`+`1999-2006`,regime_3_net_change =`1982-1998`+`2007-2019`) %>%drop_na()```::: panel-tabset#### Model Summary```{r}#| label: gom regression summary# output summary - broom.helpers and gtsummary# Best Modeltbl1 <- best_gom %>%tbl_regression() %>%add_glance_table() %>%bold_labels()tbl1```#### VIF```{r}#| label: gom vif# make a model with all of themall_parms <-lm( b ~ regime * SST + regime * GSI + regime * landings + regime * zp_small + regime * zp_large, data = gom_df)# Check VIFvif(all_parms, type ="predictor") %>%as.data.frame() %>%rownames_to_column(var ="Predictor") %>%gt() ```#### Actual vs. Fitted```{r}#| label: gom fitaugment(best_gom, gom_df_lag ) %>%# augment(best_gom, gom_df) %>% ggplot(aes(year)) +geom_line(aes(y = b, color ="Actual"), linewidth =1) +geom_line(aes(y = .fitted, color ="Predicted"), linewidth =1) +scale_color_gmri() +labs(color ="Color", y ="Spectrum Slope")```:::### Georges BankThe best supported model for Georges Bank retained two predictors: sea surface temperature anomalies (p = 0.028) and the small zooplankton index at a 2-year lag (p = 0.006). With both the SST anomalies and the lagged small zooplankton index having negative effects on size spectrum slope. Five candidate models were within a delta-AIC range of \<2. These top-models retained only bottom-up drivers as the best predictors, usually as some pair containing the small zooplankton at 0-2 year lags and either SST anomalies or the Gulf Stream Index. Suggesting that this region's size spectrum changes are most highly correlated with environmental forces and not commercial landings.```{r}#| label: gb models# Georges Bank showedgb_df_lag <- gb_df %>%mutate(b = b,blag1 = dplyr::lag(b, 1),blag2 = dplyr::lag(b, 2),zpslag1 = dplyr::lag(zp_small, 1),zpslag2 = dplyr::lag(zp_small, 2)) %>%drop_na(blag2)# fit model with all parametersall_parms <-lm(b ~ SST + GSI + landings + zp_small + zp_large + zpslag1 + zpslag2, data = gb_df_lag)# Refine the dredgeresults <-dredge(all_parms)#grab best model# subset(results, delta == 0)bmod <-get.models(results, subset = delta ==0)[[1]]``````{r}#| label: gb-top-models-table# Pull details for models with delta AIC < 2top_mods_gb <-get.models(results, subset = delta <2)# Peformance, Gets the AIC order top_perf <-map_dfr(top_mods_gb,function(x){glance(x) %>%mutate(AICc =AICc(x)) }, .id ="id") %>%arrange(AICc)# Calculate delta AICctop_perf$Delta <- top_perf$AICc -min(top_perf$AICc)# Give the ranking based on that ordertop_perf <- top_perf %>%arrange(Delta) %>%mutate(`Model ID`=row_number(),`Model ID`=str_c("Model ", `Model ID`))# Top Model Coefficients, join in the model number based on AIC ordertop_coef <-map_dfr(top_mods_gb, tidy, .id ="id") %>%left_join(select(top_perf, id, `Model ID`)) %>%select(-id)# Reshape Performance Specs to bind_row latertop_perf <-select(top_perf, `Model ID`, r2 = r.squared, Delta) %>%pivot_longer(cols =-1, names_to ="term", values_to ="estimate") %>%mutate(fill_color ="lightgray")# Clean up the data and put it into a graphgb_table_data <- top_coef %>%# Tidy up the names and set the color scheme for fillmutate(term =ifelse(term =="landings", "Commercial Landings", term),term =ifelse(term =="zp_small", "Small Zooplankton", term),term =ifelse(term =="zp_large", "Large Zooplankton", term),term =ifelse(term =="zpslag1", "Small Zooplankton 1-yr Lag", term),term =ifelse(term =="zpslag2", "Small Zooplankton 2-yr Lag", term),fill_color =ifelse(estimate<0, "#EACA00", "#00736D")) %>%# Join in the Performance Details and set Factor Ordersbind_rows(top_perf) %>%mutate(survey_area ="Georges Bank",`Model ID`=fct_rev(`Model ID`),estimate =round(estimate, 2)) %>%# Set the different types of estimates heremutate(type =case_when( term =="r2"~"Model\nPerformance", term =="Delta"~"Model\nPerformance",str_detect(term,":") ~"Regime\nInteraction",TRUE~"Fixed\nEffects"),type =factor(type, levels =c("Fixed\nEffects", "Model\nPerformance"))) ```::: panel-tabset#### Model Summary```{r}#| label: gb regression summary# output summary - brrom.helpers and gtsummary# Best Modeltbl1 <- bmod %>%tbl_regression() %>%add_glance_table() %>%bold_labels()tbl1```#### VIF```{r}#| label: gb vifall_parms <-lm(b ~ SST + GSI + landings + zp_small + zp_large + zpslag1 + zpslag2, data = gb_df_lag)vif(all_parms) %>%as.data.frame() %>%rownames_to_column(var ="Predictor") %>%rename(VIF =".") %>%arrange(VIF) %>%gt() ```#### Actual vs. Fitted```{r}#| label: gb fitaugment(bmod, gb_df_lag) %>%ggplot(aes(year)) +geom_line(aes(y = b, color ="Actual"), linewidth =1) +geom_line(aes(y = .fitted, color ="Predicted"), linewidth =1) +scale_color_gmri() +labs(color ="Color", y ="Spectrum Slope")```:::### Southern New EnglandExploratory analysis for Southern New-England was suggestive of a lack of trend in size spectrum slopes. This was further confirmed by the model selection process. The best supported model of Southern New England retained only commercial landings, however this relationship was not significant (p = 0.13) and had very low performance (r-squared = 0.06).```{r}#| label: sne models# fit model with all parametersall_parms <-lm(b ~ SST + GSI + landings + zp_small + zp_large, data = sne_df)# Refine the dredgeresults <-dredge(all_parms)#grab best model# subset(results, delta == 0)bmod <-get.models(results, subset = delta ==0)[[1]]``````{r}#| label: sne-top-models-table# Pull details for models with delta AIC < 2top_mods_sne <-get.models(results, subset = delta <2)# Peformance, Gets the AIC order top_perf <-map_dfr(top_mods_sne,function(x){glance(x) %>%mutate(AICc =AICc(x)) }, .id ="id") %>%arrange(AICc)# Calculate delta AICctop_perf$Delta <- top_perf$AICc -min(top_perf$AICc)# Give the ranking based on that ordertop_perf <- top_perf %>%arrange(Delta) %>%mutate(`Model ID`=row_number(),`Model ID`=str_c("Model ", `Model ID`))# Top Model Coefficients, join in the model number based on AIC ordertop_coef <-map_dfr(top_mods_sne, tidy, .id ="id") %>%left_join(select(top_perf, id, `Model ID`)) %>%select(-id)# Reshape Performance Specs to bind_row latertop_perf <-select(top_perf, `Model ID`, r2 = r.squared, Delta) %>%pivot_longer(cols =-1, names_to ="term", values_to ="estimate") %>%mutate(fill_color ="lightgray")# Clean up the data and put it into a graphsne_table_data <- top_coef %>%# Tidy up the names and set the color scheme for fillmutate(term =ifelse(term =="landings", "Commercial Landings", term),term =ifelse(term =="zp_small", "Small Zooplankton", term),term =ifelse(term =="zp_large", "Large Zooplankton", term),fill_color =ifelse(estimate<0, "#EACA00", "#00736D")) %>%# Join in the Performance Details and set Factor Ordersbind_rows(top_perf) %>%mutate(survey_area ="Southern New England",`Model ID`=fct_rev(`Model ID`),estimate =round(estimate, 2)) %>%# Set the different types of estimates heremutate(type =case_when( term =="r2"~"Model\nPerformance", term =="Delta"~"Model\nPerformance",str_detect(term,":") ~"Regime\nInteraction",TRUE~"Fixed\nEffects"),type =factor(type, levels =c("Fixed\nEffects", "Model\nPerformance"))) ```::: panel-tabset#### Model Summary```{r}#| label: sne regression summary# output summary - brrom.helpers and gtsummary# Best Modeltbl1 <- bmod %>%tbl_regression() %>%add_glance_table() %>%bold_labels()tbl1```#### VIF```{r}#| label: sne vifall_parms <-lm( b ~ SST + GSI + landings + zp_small + zp_large, data = sne_df)vif(all_parms) %>%as.data.frame() %>%rownames_to_column(var ="Predictor") %>%rename(VIF =".") %>%arrange(VIF) %>%arrange(VIF) %>%gt() ```#### Actual vs. Fitted```{r}#| label: sne fitaugment(bmod, sne_df) %>%ggplot(aes(year)) +geom_line(aes(y = b, color ="Actual"), linewidth =1) +geom_line(aes(y = .fitted, color ="Predicted"), linewidth =1) +scale_color_gmri() +labs(color ="Color", y ="Spectrum Slope")```:::### Mid-Atlantic BightThree top models were selected for the Mid-atlantic Bight region, all with non-stationary predictors. The best supported model showed that increases in the small zooplankton index had a negative impact on spectrum slope (p = 0.020). This result was present in the other top candidates, which each also included a negative correlation to large zooplankton or with commercial landings. Model performance was low among the top models (r-squared 0.14-0.16).```{r}#| label: mab models# fit model with all parametersall_parms <-lm(b ~ SST + GSI + landings + zp_small + zp_large, data = mab_df)# Refine the dredgeresults <-dredge(all_parms)#grab best modelbmod <-get.models(results, subset = delta ==0)[[1]]``````{r}#| label: mab-top-models-table# Pull details for models with delta AIC < 2top_mods_mab <-get.models(results, subset = delta <2)# Peformance, Gets the AIC order top_perf <-map_dfr(top_mods_mab, function(x){glance(x) %>%mutate(AICc =AICc(x)) }, .id ="id") %>%arrange(AICc)# Calculate delta AICctop_perf$Delta <- top_perf$AICc -min(top_perf$AICc)# Give the ranking based on that ordertop_perf <- top_perf %>%arrange(Delta) %>%mutate(`Model ID`=row_number(),`Model ID`=str_c("Model ", `Model ID`))# Top Model Coefficients, join in the model number based on AIC ordertop_coef <-map_dfr(top_mods_mab, tidy, .id ="id") %>%left_join(select(top_perf, id, `Model ID`)) %>%select(-id)# Reshape Performance Specs to bind_row latertop_perf <-select(top_perf, `Model ID`, r2 = r.squared, Delta) %>%pivot_longer(cols =-1, names_to ="term", values_to ="estimate") %>%mutate(fill_color ="lightgray")# Clean up the data and put it into a graphmab_table_data <- top_coef %>%# Tidy up the names and set the color scheme for fillmutate(term =ifelse(term =="landings", "Commercial Landings", term),term =ifelse(term =="zp_small", "Small Zooplankton", term),term =ifelse(term =="zp_large", "Large Zooplankton", term),fill_color =ifelse(estimate<0, "#EACA00", "#00736D"),fill_color =ifelse(term =="(Intercept)", "lightgray", fill_color)) %>%# Join in the Performance Details and set Factor Ordersbind_rows(top_perf) %>%mutate(survey_area ="Mid-Atlantic Bight",`Model ID`=fct_rev(`Model ID`),estimate =round(estimate, 2)) %>%# Set the different types of estimates heremutate(type =case_when( term =="r2"~"Model\nPerformance", term =="Delta"~"Model\nPerformance",str_detect(term,":") ~"Regime\nInteraction",TRUE~"Fixed\nEffects"),type =factor(type, levels =c("Fixed\nEffects", "Model\nPerformance"))) ```::: panel-tabset#### Model Summary```{r}#| label: mab regression summary# output summary - brrom.helpers and gtsummary# Best Modeltbl1 <- bmod %>%tbl_regression() %>%add_glance_table() %>%bold_labels()tbl1```#### VIF```{r}#| label: mab vifall_parms <-lm(b ~ SST + GSI + landings + zp_small + zp_large, data = mab_df)vif(all_parms) %>%as.data.frame() %>%rownames_to_column(var ="Predictor") %>%rename(VIF =".") %>%arrange(VIF) %>%gt() ```#### Actual vs. Fitted```{r}#| label: mab fitaugment(bmod, mab_df ) %>%ggplot(aes(year)) +geom_line(aes(y = b, color ="Actual"), linewidth =1) +geom_line(aes(y = .fitted, color ="Predicted"), linewidth =1) +scale_color_gmri() +labs(color ="Color", y ="Spectrum Slope")```::::::### Model Selection Results```{r}#| label: model selection table#| fig-height: 10#| fig-width: 8bind_rows(list( gom_table_data, sne_table_data, mab_table_data, gb_table_data)) %>%mutate(term =ifelse(term =="Delta", "Delta AICc", term),survey_area =factor(survey_area, levels = area_levels_long),`Model ID`=factor(`Model ID`, levels =str_c("Model ", c(5:1))))%>%ggplot( aes(term, `Model ID`)) +geom_tile(aes(fill =I(fill_color)), color ="gray20", linewidth =0.2, alpha =0.75) +geom_text(aes(label = estimate)) +facet_grid(survey_area ~ type, scales ="free", space ="free") +theme(axis.text.x =element_text(angle =45, hjust =1, vjust =1)) +labs(x ="Model Predictor",y ="",title ="Spectrum Slope Driver Exploration",subtitle ="Interaction terms should be added to the corresponding fixed effect to determine net-influence." )```# Discussion### Potential Drivers Timeseries:The following panels show the historical changes in each of the drivers. Landings have been scaled by average total landings within each region across all years. SST and GSI have not been scaled. This is different from how they are implemented in the regression analyses, when the landings were scaled over the 1982-2019 period.# Supplemental Materials::: panel-tabset## Species Functional Groups```{r}#| label: tbl-functional-groups#| tbl-cap: "Common and scientific names for the species that constitute each functional group used in our analyses. X markers are used to indicate which regions each species has been caught in the data."# Tidy up a key of what is labeled whatspec_key <- catch_size_bins %>%distinct(comname, scientific_name, hare_group) %>%mutate(hare_group =ifelse(hare_group %in%c("elasmobranch", "groundfish"), hare_group, str_c(hare_group)),hare_group =ifelse(is.na(hare_group), "Not Classified", hare_group),hare_group =str_to_title(hare_group),hare_group =fct_drop(hare_group),hare_group =factor(hare_group, levels =c("Coastal", "Diadromous", "Elasmobranch", "Groundfish", "Pelagic", "Reef", "Not Classified"))) %>%group_by(hare_group) %>%mutate(n =str_c("(",n(),")")) %>%ungroup()# Reformat the data to check presence absencecatch_size_bins %>%distinct(comname, survey_area) %>%mutate(Presence ="X") %>%complete(comname, survey_area) %>%left_join(spec_key, by ="comname") %>%mutate(comname =str_to_title(comname),Presence =ifelse(is.na(Presence) ==FALSE, "X", ""),Presence =ifelse(is.na(Presence) ==TRUE, "", Presence)) %>%left_join(area_df, by ="survey_area") %>%select(hare_group, n, comname, scientific_name, area, Presence) %>%pivot_wider(names_from = area, values_from = Presence) %>%arrange(hare_group, comname) %>%group_by(hare_group,n) %>%gt() %>% gt::tab_header(title =md("**Functional Group Assignments and Regional Presence/Absence**")) %>%tab_stubhead(label =md("**Functional Group**")) %>%cols_label(comname =md("*Common Name*"),scientific_name =md("*Scientific Name*"),`Gulf of Maine`=md("*Gulf of Maine*"),`Georges Bank`=md("*Georges Bank*"),`Southern New England`=md("*Southern New England*"),`Mid-Atlantic Bight`=md("*Mid-Atlantic Bight*") ) %>%tab_style(style =list(cell_text(weight ="bold")),locations =cells_row_groups()) %>% gt::tab_source_note(source_note =md("*Functional group assignments adapted from Hare et al. 2010*"))```## GARFO Landings Summary```{r}#| label: garfo-landings-summary# Add the labels into the landings data and remove what we don't need there:landings %>%filter(between(year, 1960, 2019),!str_detect(sppname, "CONFIDENTIAL")) %>%mutate(decade =floor_decade(year),sppname =str_to_title(sppname)) %>%group_by(survey_area, decade, sppname) %>%summarise(avg_landings_lb =mean(`landed lbs`, na.rm = T),across(.cols =c(landings_lb =`landed lbs`, value = value), .fns = sum, .names ="total_{.col}"),.groups ="drop") %>%group_by(survey_area, decade) %>%slice_max(order_by = total_landings_lb, n =3) %>%gt() %>% gt::tab_header(title =md("**Top Commercial Fisheries Landings of Northeastern US (by weight)**")) %>%tab_stubhead(label =md("*Harvest Region*")) %>%fmt_number(columns =c(avg_landings_lb, total_landings_lb, total_value),use_seps = T, sep_mark =",",suffixing = T) %>%tab_style(style =list(cell_text(weight ="bold")),locations =cells_row_groups()) %>%cols_label(decade =md("*Decade*"),sppname =md(""),avg_landings_lb =md("*Avg. Annual Landings (lb.)*"),total_landings_lb =md("*Total Landings (lb.)*"),total_value =md("*Total Value ($)*")) %>% gt::tab_source_note(source_note =md("*Landings data obtained from the Greater Atlantic Regional Fishing Office (GARFO)*"))```:::### Single Driver Correlations```{r}#| label: period-correlations-processing# Run Correlations for Each period and for the entire perioddriver_period_correlations <-map(driver_period_list, function(drivers){#------ debug copy-paste line# Step 1. Get the correlation matrix (data is pre-scaled) corr_matrix <-rcorr(as.matrix(drivers))# R-squares rho_df <-as.data.frame(corr_matrix$r) %>%rownames_to_column(var ="var1") %>%pivot_longer(cols =-var1, names_to ="var2", values_to ="r2")# Significance Dataframe sig_df <-as.data.frame(corr_matrix$P) %>%rownames_to_column(var ="var1") %>%pivot_longer(cols =-var1, names_to ="var2", values_to ="pval")# Join corr_both <-left_join(rho_df, sig_df, by =c("var1", "var2")) # Step 2. Tidy Up for ggplot# Make correlation matrix & significance a dataframe for GGplot# Goal: Flag the different Functional Group size changes somehow# Mega Cleanup Code Chunk# This first step adds metadata, but mostly creates a way to filter out correlations to be in the direction we want. no sst as response variables for example# GOAL: All the drivers in one column, and all the responses in another, order doesn't matter corr_tidy <- corr_both %>%mutate(# 1.These flag whether or not the matrix variables in each column are the independent or dependent variablesis_var1_driver =ifelse(str_detect(var1, "landings|index|anom|avg|zp"), T, F),is_var2_response =ifelse(str_detect(var2, "slope|int"), T, F),# 2. Column to Flag what type of driver it is:driver_type =case_when(str_detect(var1, "landings") ~"Top-Down",str_detect(var1, "stream|anom|zp") ~"Bottom-Up",str_detect(var1, "length") ~"Avg. Length",str_detect(var1, "weight") ~"Avg. Weight"),# 3. Column to Flag what area the driver data is fromdriver_area =case_when(str_detect(var1, "All") ~"Northeast Shelf",str_detect(var1, "Georges") ~"Georges Bank",str_detect(var1, "Gulf") ~"Gulf of Maine",str_detect(var1, "Southern") ~"Southern New England",str_detect(var1, "Mid-Atlantic") ~"Mid-Atlantic Bight"),# 4. Set Factor Levels for plotdriver_area =factor( driver_area, levels =c("All", "Gulf of Maine", "Georges Bank", "Southern New England", "Mid-Atlantic Bight")),# 5. Shorthand Names for the Spectra Parameters - Responseresponse_short =case_when(str_detect(var2, "int") ~"Spectra Intercept",str_detect(var2, "isd") ~"ISD Exponent",str_detect(var2, "slope") ~"Spectra Slope"),# 6. These are the areas of the driversresponse_area =case_when(str_detect(var2, "All") ~"All",str_detect(var2, "Georges") ~"Georges Bank",str_detect(var2, "Gulf") ~"Gulf of Maine",str_detect(var2, "Southern") ~"Southern New England",str_detect(var2, "Mid-Atlantic") ~"Mid-Atlantic Bight"),# 7. Set levels for the driver areasresponse_area =factor( response_area, levels =c("All", "Gulf of Maine", "Georges Bank", "Southern New England", "Mid-Atlantic Bight")),# 8. Flag when the areas are the same or not:area_match = driver_area == response_area )# Now we filter rows out to focus in on:# 1. one direction driver & response# 2. response and driver in same area, unless its GSIcorr_tidy_pairs <- corr_tidy %>%filter( is_var2_response, is_var1_driver, area_match |str_detect(var1, "index")) # From here it is just text formatting for plots:# Because we are faceting we can have repeated names to simplify thingscorr_tidy_pairs <- corr_tidy_pairs %>%mutate(var1 =ifelse(str_detect(var1, "landings"), "Commercial_Landings", var1),var1 =ifelse(str_detect(var1, "sst"), "SST_Anomalies", var1),var1 =ifelse(str_detect(var1, "gulf_stream_index"), "Gulf Stream Index", var1),var1 =ifelse(str_detect(var1, "zp_small"), "Small Zooplankton", var1),var1 =ifelse(str_detect(var1, "zp_large"), "Large Zooplankton", var1),var1 =str_replace_all(var1, "_", " "),# Make a functional group columnhare_group =case_when(str_detect(var1, "elasmobranch") ~"Elasmobranch",str_detect(var1, "pelagic") ~"Pelagic",str_detect(var1, "coastal") ~"Coastal",str_detect(var1, "diadromous") ~"Diadromous",str_detect(var1, "groundfish") ~"Groundfish",str_detect(var1, "reef") ~"Reef",TRUE~""),# Add the functional group column ahead of the average length and weight columnsvar1 =ifelse(str_detect(var1, "avg"), hare_group, var1),# Set The Factor Levels groupsvar1 =factor( var1,levels =c("Gulf Stream Index","SST Anomalies","Small Zooplankton","Large Zooplankton","Groundfish","Elasmobranch","Coastal","Diadromous","Pelagic","Reef","Commercial Landings")),# Flag Significance Levelssignif =ifelse(pval <0.05, "*", ""))return(corr_tidy_pairs)#------- })```::: panel-tabset#### All Years```{r}#| label: correlations-all#| fig-height: 6#| fig-width: 8period <-"1982-2019"plot_period_correlations(driver_period_correlations[[period]], period)```#### 1982-1998```{r}#| label: correlations-80s#| fig-height: 6#| fig-width: 8period <-"1982-1998"plot_period_correlations(driver_period_correlations[[period]], period)```#### 1999-2006```{r}#| label: correlations-90s#| fig-height: 6#| fig-width: 8period <-"1999-2006"plot_period_correlations(driver_period_correlations[[period]], period)```#### 2007-2019```{r}#| label: correlations-20s#| fig-height: 6#| fig-width: 8period <-"2007-2019"plot_period_correlations(driver_period_correlations[[period]], period)```:::```{r}#| label: weisberg-synthesis#| eval: false# Stuff for Sara:# 1. Plot of Gulf of Maine Slopes, no gray linessara_plot <- region_indices %>%filter(survey_area =="GoM") %>%ggplot(aes(yr, b, group = area_titles)) +# geom_line(aes(group = area_titles), linewidth = 1) +geom_point(size =1) +geom_smooth(method ='loess', formula ='y ~ x', color ="orange") +#geom_pointrange(aes(group = area_titles, ymin = confMin, ymax = confMax), size = 0.25) +scale_color_gmri() +facet_wrap(~area_titles, ncol =1) +labs(x ="Year", y ="Abundance Size Spectrum Slope (b)", title ="Exponent of Abundance Size Spectra")# The decline in weight of groundfish to go with:# Average individual weightfunctional_group_size %>%filter(survey_area =="GoM", hare_group %in%c("groundfish", "elasmobranch")) %>%ggplot(aes(Year, mean_wt_kg)) +geom_point(size =1) +geom_smooth(color ="orange", method ="loess", linewidth =1) +facet_grid(hare_group~area_titles, scales ="free_y") +theme(axis.text =element_text(size =8)) +labs(title ="Average Weight",y ="Weight (kg)",color ="Region")# How much biomass is just gf and elsmo:# Plot abundance by region and Groupfgroup_area$weight_bins %>%filter(survey_area =="GoM") %>%ggplot(aes(Year, strat_lwbio_mill, fill = hare_group)) +geom_col(color ="gray30", linewidth =0.1, position ="fill") +facet_grid(.~survey_area) +scale_fill_gmri() +scale_y_continuous(labels =percent_format()) +labs(y ="Fraction of Biomass", x =NULL, fill ="Functional Group") +theme(legend.position ="bottom",axis.text.x =element_text(angle =45, hjust =1))# Table of averages over different periods for all of them:region_indices %>%filter(yr %in%c(1970:2000)) %>%mutate(year_range ="1970-2000") %>%group_by(year_range, area_titles) %>%summarise(avg_slope =mean(b, na.rm = T),sd_slope =sd(b, na.rm = T) )```# References
